…arch#35756)

Lower the model_catalog disk-cache TTL from 24h to 1h so freshly
published model-catalog.json deploys reach the picker within an hour
instead of up to a day. The picker now refetches on the next
`hermes model` / `/model` once the cache is older than 1h; younger
than 1h still serves the cache (no network hit), and network failures
still fall back to the stale copy.

- DEFAULT_TTL_HOURS 24 -> 1 (model_catalog.py)
- DEFAULT_CONFIG model_catalog.ttl_hours 24 -> 1, _config_version 24 -> 25
- migration v24->25 rewrites a stale ttl_hours:24 to 1, preserving any
  custom value the user set

E2E: verified >1h refetches / <1h skips, and migration rewrites 24->1
while preserving a custom 6.

)

* fix: detect pyproject.toml / __init__.py version drift in hermes doctor (#35142)

A git conflict resolution (reset --hard or merge) can revert
hermes_cli/__init__.py to a stale __version__ while pyproject.toml stays
current, so 'hermes --version' silently reports the wrong version. Nothing
cross-checked the two files.

Add a version-consistency check to the doctor 'Python Environment' section:
reads the [project] version from pyproject.toml and compares it to
hermes_cli.__version__. Reports OK when they match, fails with a re-sync
hint when they drift, and is a silent no-op for installed wheels where
pyproject.toml isn't present.

Closes #35070

* fix: surface /agents nudge while delegate_task is in-flight (TUI + CLI)

The subagent spawn-observability overlay added a `(/agents)` hint, but
only on the standalone "Spawn tree" panel, gated behind `!inlineDelegateKey`
— it never showed for a single delegate_task call, and only appeared once
subagents had already registered. A nudge that arrives at the end (or only
after spawn) is useless for the actual goal: letting users open the live
monitor *while* delegation is running.

Surface it the moment delegation starts, on both surfaces:

TUI (ui-tui/src/components/thinking.tsx)
- Show `(/agents)` on any "Delegate Task" tool group as soon as it appears
  (in-flight, before any subagent registers), not gated on subagents
  already existing. Same `startsWith('Delegate Task')` predicate already
  used for delegateGroups.

CLI (agent/tool_executor.py)
- Append `· /agents to monitor` to the delegate spinner label, which is
  displayed for the full duration of the delegate_task call. The previous
  attempt put the hint on the completion line (get_cute_tool_message),
  which only renders after the call finishes — reverted.

TUI tsc clean (pre-existing execFileNoThrow type errors unrelated);
subagentTree 35/35; display.py reverted to upstream.

* fix(vision): fail fast on non-retryable image download errors (#35221)

_download_image() wrapped every download attempt in a blanket
`except Exception` and retried 3x with 2s/4s/8s backoff regardless of
cause. A 404/403 image URL would never resolve on retry, so it just
burned up to 6s of wall-clock + extra GETs before failing — inflating
latency for a deterministic failure (issue #32296, umbrella #35114).

Add _is_retryable_download_error(): 4xx client errors (except 429),
website-policy PermissionError, and too-large/SSRF ValueError now raise
on the first attempt. 429, 5xx, and unclassified network errors stay
retryable. Removed the now-unreachable fall-through branch since the
loop always returns on success or re-raises on the final/terminal attempt.

* fix(tui): use base64 encoding for PowerShell clipboard writes to preserve UTF-8

When writing text to the clipboard via PowerShell (WSL2 and native Windows),
the previous implementation piped text through stdin using `Set-Clipboard
-Value $input`. PowerShell reads stdin using the Windows system's default
ANSI code page (e.g. CP936 for Chinese Windows), causing all non-ASCII
characters (CJK, emoji, accented) to become garbled.

Fix: encode the text as base64 in Node.js and pass it as a command argument.
PowerShell decodes it from base64 using explicit UTF-8, bypassing the code
page issue entirely.

Fixes #35107

* refactor(tui): simplify base64 clipboard write to a stdin flag

The per-entry psScript callback was identical for every PowerShell entry,
so the function-valued union member added structure without behavior. Collapse
WriteCmd to a plain stdin boolean and apply the one shared base64 script in the
write loop. Document the CP936 root cause inline.

Co-authored-by: BROCCOLO1D <279959838+BROCCOLO1D@users.noreply.github.com>

* fix(kanban): rebuild legacy TEXT-PK tables to INTEGER AUTOINCREMENT on open

Legacy kanban boards (pre-AUTOINCREMENT schema) crashed the gateway
notifier on every tick — int(None) on a NULL id in unseen_events_for_sub
— silently losing all kanban notifications. CREATE TABLE IF NOT EXISTS
skips existing tables regardless of schema and _add_column_if_missing
only adds columns, so neither could fix a drifted primary-key type.

_rebuild_drifted_tables() detects the legacy shape via PRAGMA table_info
and rebuilds task_events/task_comments/task_runs (TEXT PK -> INTEGER
AUTOINCREMENT) and kanban_notify_subs.last_event_id (TEXT/NULL -> INTEGER
NOT NULL DEFAULT 0), preserving data. The whole pass is one transaction
so an interruption can't leave a table half-renamed, and recreates every
index DROP TABLE would otherwise take down (including idx_events_run).

Co-authored-by: liuhao1024 <liuhao1024@users.noreply.github.com>

* fix(kanban): prevent infinite retry loop when worker exhausts iteration budget

recompute_ready() previously reset consecutive_failures to 0 when
auto-recovering a blocked task.  This defeated the circuit-breaker:
a task that repeatedly exhausted its iteration budget would cycle
forever (block → auto-recover with counter=0 → respawn → budget
exhausted → block → …) with no signal to the operator.

Fix: don't auto-recover tasks whose consecutive_failures has reached
the effective failure limit (per-task max_retries or
DEFAULT_FAILURE_LIMIT).  The counter is also preserved across
recovery so the breaker can accumulate across cycles.

Fixes #35072

* fix(kanban): align recompute_ready guard with breaker's configured failure_limit

Follow-up to the budget-exhaustion recovery fix. recompute_ready's
new circuit-breaker guard resolved its effective limit from per-task
max_retries -> DEFAULT_FAILURE_LIMIT, skipping the dispatcher's
configured kanban.failure_limit. _record_task_failure resolves
max_retries -> failure_limit(config) -> DEFAULT, so the two disagreed
whenever an operator set kanban.failure_limit != 2:

- config > 2: a task could get stuck at DEFAULT(2) before reaching its
  allowed retry count.
- config < 2: a task the breaker already blocked could be auto-recovered
  back to ready, defeating the stricter limit.

Thread the dispatcher's failure_limit through dispatch_once into
recompute_ready so the guard and the breaker share one resolution order.
Updated test_circuit_breaker_block_still_auto_promotes (it asserted a
failures=5 block auto-recovers and resets the counter — that's the
pre-#35072 behavior the loop fix removes); it now exercises a below-limit
transient block, with the at-limit case covered in test_kanban_db.py.
Added two tests for the config-tier and per-task override resolution.

* fix(update): export launcher virtualenv to uv

* feat(model-picker): group multi-endpoint providers under one row (#35227)

* Inspired by Claude Code: /compress here [N] — boundary-aware 'summarize up to here'

Adds a user-chosen compression boundary to the existing /compress command.
/compress here [N] summarizes everything except the most recent N exchanges
(default 2), which are preserved verbatim — letting the user pick the
compression boundary instead of relying on the automatic token-budget heuristic.

Inspired by Claude Code's Rewind 'Summarize up to here' action (v2.1.139,
Week 20, May 2026): https://code.claude.com/docs/en/whats-new/2026-w20

- hermes_cli/partial_compress.py: pure split/parse helpers + seam-alternation
  guard (shared by CLI and gateway).
- cli.py / gateway/run.py: route 'here [N]' / '--keep N' to partial compression;
  compress only the head, re-append the verbatim tail through the seam guard.
- Preserves message-flow role alternation (seam guard merges any illegal
  user->user / assistant->assistant adjacency).
- Reuses the existing _compress_context session-rotation/lock machinery — no
  changes to the compression core.
- Bare /compress (full) and /compress <focus> behavior unchanged.

Tests: 12 helper unit tests + 5 CLI integration tests + E2E (interleaved
tool-call transcript, degenerate/multimodal seams, real handler path).

* feat(model-picker): group multi-endpoint providers under one row

The interactive provider pickers (hermes model, setup wizard, Telegram
/model) listed every provider slug flat, so vendors with several endpoints
(Kimi/Moonshot, MiniMax, xAI Grok, Google Gemini, OpenAI, OpenCode, GitHub
Copilot) each occupied multiple top-level rows. Now related slugs fold into
one top-level row that drills down to the specific endpoint.

- models.py: add PROVIDER_GROUPS table + group_providers() fold (display
  only — CANONICAL_PROVIDERS, slugs, --provider, /model <provider:model>
  all unchanged and individually addressable).
- hermes model (main.py): group rows drill into a member sub-picker, then
  dispatch to the existing _model_flow_* unchanged. setup wizard inherits it.
- Telegram /model: new mpg:<group> callback expands to member mp:<slug>
  buttons; single authenticated member degrades to a direct button.
- Grouping is the single shared fold across all three surfaces.

Validation: 163 targeted tests pass; E2E confirms group->member->model
resolves to the correct concrete slug for all families.

* fix(packaging): include mcp_serve in py-modules so hermes mcp serve works on pip installs

mcp_serve.py was missing from the setuptools py-modules list, causing
hermes mcp serve to crash with ModuleNotFoundError on standard pip installs.

Fixes #34871

* fix(skills): fix transaction ordering in reset_bundled_skill and handle read-only files in rmtree

Two related bugs in tools/skills_sync.py affecting Nix-store and
immutable-package installs:

**#34972 — reset_bundled_skill corrupts manifest on rmtree failure:**
The function deleted the manifest entry BEFORE attempting rmtree. If
rmtree failed (read-only files from Nix store), the function returned
early — leaving the skill in a manifest-less limbo state where future
syncs silently skip it forever.

Fix: reorder steps — attempt rmtree FIRST, only delete manifest entry
after rmtree succeeds. If rmtree fails, nothing is changed.

**#34860 — stale .bak directories after sync:**
sync_skills() called shutil.rmtree(backup, ignore_errors=True) which
silently failed on read-only files, leaving persistent .bak dirs.

Fix: add _rmtree_writable() helper that makes files writable via an
onerror callback before retrying removal. Used in both sync_skills()
backup cleanup and reset_bundled_skill().

Fixes #34972
Fixes #34860

* fix(skills): make _rmtree_writable handle read-only directories, not just files

The cherry-picked fix's onerror handler chmod'd only the failing path, but
unlinking a child requires write permission on its PARENT directory. On a true
Nix-store copy (r-xr-xr-x dirs + files) rmtree still failed. Now chmod the
parent dir as well before retrying.

Also rewrites the regression test: the original asserted the helper FAILS on a
read-only dir (documenting the limitation), which is the wrong success criterion.
Split into two tests — restore succeeds on a full read-only tree (real Nix case),
and manifest is preserved when removal genuinely cannot proceed (monkeypatched).

* test(skills): assert restore via synced[copied], not manifest re-read

The hermetic CI env (slice 4/6) redirects HERMES_HOME, so a post-restore
_read_manifest() can resolve to an empty/redirected manifest path and return
{}. Assert on sync_skills's in-memory return value (synced["copied"]) instead,
which is the resilient signal that the skill was re-copied and is no longer in
limbo.

* fix(file-tools): make write_file/patch atomic (temp-file + rename) (#35252)

* Inspired by Claude Code: /compress here [N] — boundary-aware 'summarize up to here'

Adds a user-chosen compression boundary to the existing /compress command.
/compress here [N] summarizes everything except the most recent N exchanges
(default 2), which are preserved verbatim — letting the user pick the
compression boundary instead of relying on the automatic token-budget heuristic.

Inspired by Claude Code's Rewind 'Summarize up to here' action (v2.1.139,
Week 20, May 2026): https://code.claude.com/docs/en/whats-new/2026-w20

- hermes_cli/partial_compress.py: pure split/parse helpers + seam-alternation
  guard (shared by CLI and gateway).
- cli.py / gateway/run.py: route 'here [N]' / '--keep N' to partial compression;
  compress only the head, re-append the verbatim tail through the seam guard.
- Preserves message-flow role alternation (seam guard merges any illegal
  user->user / assistant->assistant adjacency).
- Reuses the existing _compress_context session-rotation/lock machinery — no
  changes to the compression core.
- Bare /compress (full) and /compress <focus> behavior unchanged.

Tests: 12 helper unit tests + 5 CLI integration tests + E2E (interleaved
tool-call transcript, degenerate/multimodal seams, real handler path).

* fix(file-tools): make write_file/patch atomic (temp-file + rename)

write_file streamed content straight into the target via `cat > path`, so
a crash, SIGKILL, or truncated pipe mid-write left the file half-written
and corrupt. patch_replace routes through write_file, so it shared the flaw.

Now writes stream into a temp file in the SAME directory and `mv` it over
the target — a real same-filesystem rename, which is atomic on POSIX and on
every terminal backend (local/docker/ssh/modal). A failed write leaves the
original byte-intact and leaks no temp file. The existing file's mode is
preserved across the swap (stat + chmod, GNU/BSD), and content still rides
stdin so there's no ARG_MAX limit. A trap cleans the temp on any error path.

Tests: added TestAtomicWrite (real LocalEnvironment, no mocks) covering
inode-change-on-overwrite, mode preservation, failed-write-leaves-original,
no-temp-leak, special chars, and patch routing. Updated two mocks in
test_file_operations.py that keyed on the literal `cat >` write command to
key on the stdin_data behavioral signal instead. 200 file-tool tests green.

* fix(cli): use `uv tool upgrade` when Hermes is a uv tool install (#29700)

Hermes installed via `uv tool install hermes-agent` lives outside any
venv. `_cmd_update_pip` previously ran `uv pip install --upgrade`, which
errors with `No virtual environment found; run uv venv ...`. The user
hits this on the very first `hermes update` after a standard
non-`--system` install with `uv` on PATH.

Add `is_uv_tool_install()` in `hermes_cli/config.py`: fast path inspects
`sys.prefix` for the standard `uv/tools/hermes-agent/` layout, falls
back to `uv tool list` for non-standard prefixes. Both the
user-facing `recommended_update_command_for_method("pip")` string and
the actual subprocess invocation in `_cmd_update_pip` now switch to
`uv tool upgrade hermes-agent` when detected. Non-tool installs and the
no-`uv` fallback keep their existing commands unchanged.

* fix(cli): restrict uv-tool-install detection to running interpreter

Copilot review on PR #29703 flagged two issues with the `uv tool list`
fallback in `is_uv_tool_install`:

1. False positive: `uv tool list` returns the *machine*'s installed
   tools, not the active install. A regular pip/venv Hermes on a host
   that also has `uv tool install hermes-agent` available would be
   misclassified as a uv-tool install, and `hermes update` would
   upgrade the wrong copy.

2. Overhead: the subprocess call (up to a 15s timeout) was triggered
   even from `recommended_update_command_for_method`, which just
   computes a display string.

Restrict detection to properties of the running interpreter
(`sys.prefix` and `sys.executable` — both can carry the uv-tool layout
marker depending on entry point). Drop the `uv tool list` fallback and
the `uv_path` parameter entirely. `_cmd_update_pip` now also surfaces a
clear hint when the runtime looks like a uv-tool install but `uv` is
missing from PATH, instead of silently falling back to `python -m pip`.

* fix(update): handle pipx installs + --system fallback in _cmd_update_pip

Extends the uv-tool detection (briandevans, #29703) to cover the
remaining no-venv install layouts that hit the same uv 'No virtual
environment found' error:

- pipx-managed installs (sys.prefix under .../pipx/...) -> 'pipx upgrade',
  matching scripts/auto-update.sh (pipx-detection idea from
  inchargeautomation-lab, #29852)
- bare pip outside any venv -> 'uv pip install --system --upgrade'
- venv (launcher shim) keeps the VIRTUAL_ENV overlay from #35224 and never
  gets --system, so the install always targets the venv, not system Python

The four branches are mutually exclusive; VIRTUAL_ENV is exported only for
the uv-pip-in-venv path (uv tool / pipx upgrade ignore it).

Co-authored-by: Joshua Kimbrell <incharge.automation@gmail.com>

* chore(release): map inchargeautomation-lab author email

* fix(mcp): reap stdio MCP grandchildren via process-group signal

The orphan reaper for stdio MCP subprocesses only tracked the direct child
PID spawned by ``stdio_client`` (e.g. ``openclaw mcp serve``). When that
wrapper itself spawned a helper (``claude mcp serve``) and then exited, the
helper reparented to ``systemd --user`` and survived shutdown.

The MCP SDK already spawns stdio children with ``start_new_session=True``,
so the wrapper is its own pgroup leader and same-pgroup descendants are
reachable via ``killpg``. Capture the pgid at spawn time and reap via
``killpg(pgid, sig)`` so reparented grandchildren are reaped alongside the
direct child, even after the wrapper itself exits. Falls back to per-pid
``os.kill`` on Windows or when no pgid was recorded.

Fixes part 2 (orphan ``claude mcp serve``) of #23799. Part 1 (per-invocation
respawn) was confirmed by the reporter to be an environmental artifact, not
a code bug.

* test(mcp): import os and pytest in test_mcp_stability

The salvaged grandchild-reaping tests reference os.getpgid/os.killpg and
pytest.mark/skip/importorskip directly, but the file only imported asyncio,
signal, and unittest.mock. Add the missing imports so collection succeeds
on current main.

* fix(lsp): detect Windows wrapper binaries in installer probes

* fix(lsp): handle Windows .cmd shims in LSP process spawn

asyncio.create_subprocess_exec cannot run .cmd/.bat files on Windows
because CreateProcess expects a valid PE executable. npm-installed LSP
servers (intelephense, typescript-language-server, etc.) ship as .cmd
shims on Windows, causing WinError 193 on spawn.

Detect .cmd/.bat extensions and wrap with cmd.exe /c before spawning.
Gated behind sys.platform == 'win32' — no code path changes elsewhere.

Fixes #34864

* chore(release): map tuancookiez-hub for #34865 salvage

* fix(state): persist mid-session model switch to database

When a user switches models mid-session via /model, the gateway updates
the in-memory agent and session overrides, but the database was never
updated. The COALESCE(model, ?) in update_token_counts() only fills NULL
values, so the dashboard always showed the original model.

Fix: Add SessionDB.update_session_model() that unconditionally sets the
model column, and call it from both the interactive picker and direct
/model command paths in the gateway.

Fixes #34850

* test(state): cover update_session_model overwrite + getattr-guard text path

Follow-up to LengR's #35181 salvage:
- gateway text-path uses getattr(self, '_session_db', None) to match the
  picker callback path (defensive for object.__new__() gateway test pattern).
- add SessionDB.update_session_model test asserting it overwrites the
  COALESCE-pinned model and survives subsequent token updates (#34850).

* fix(run_agent): gate concurrent checkpoint preflight on block_result (fixes #34827)

In the concurrent tool-execution path, checkpoint preflight (write_file,
patch, destructive terminal) fired BEFORE plugin guardrail block_result
was computed. A blocked write_file could still dirty checkpoint state
(doc_modified_this_turn, _last_write_file_call_id, turn_counter).

Move checkpoint preflight to AFTER block_result computation, gated on
`if block_result is None:` — matching the invariant the sequential path
already enforces.

* fix(google-workspace): handle Gmail header casing case-insensitively

Normalize Gmail API message header names to lowercase before lookup so
gmail get/search/reply populate to/subject/from regardless of the casing
the message was stored with. Emit conventional MIME header casing
(To/Subject/Cc/From) on send and reply.

Fixes #34806

Co-authored-by: Donovan Yohan <donovan-yohan@users.noreply.github.com>

* fix(update/windows): robustly exclude launcher-shim ancestors from concurrent check (#35257)

hermes update on Windows still aborted with 'Another hermes.exe is running',
listing its own launcher shim(s) as concurrent instances (issues #29341,
#34795). The distlib Scripts\hermes.exe launcher spawns python.exe and waits;
detection runs in the python child, so the launcher shim shows up in
process_iter.

The prior fix walked the ancestor chain with per-hop current.parent() inside
'except: break' — the first psutil AccessDenied/NoSuchProcess (common on
Windows across session/elevation boundaries) bailed the walk early, leaving
the launcher in the candidate set and re-triggering the false positive.

- Switch to proc.parents() (whole ancestor list in one call), evaluate each
  ancestor independently so one unreadable hop never strands the launcher.
- Only exclude ancestors whose exe is itself a shim, so a genuine second
  hermes.exe under a non-Hermes parent (Desktop backend child) is still flagged.
- Message now prints a copy-pasteable 'taskkill /PID … /F' for the exact stale
  PIDs so a user who already closed everything can self-remediate.

Conservative shim-only ancestor approach credited to the parallel attempts in
PRs #29358 (xxxigm) and #31808 (jquesnelle).

* fix(auxiliary): pass base_url/api_key/api_mode through set_runtime_main for custom providers

When a user configures a custom: provider (e.g. custom:openclaw-router),
set_runtime_main() only stored provider and model in process-local globals.
_resolve_auto() then had no base_url or api_key for the custom endpoint,
causing Step 1 to fail and auxiliary tasks (approval, compression, title
generation) to fall through to the aggregator chain and route to wrong
providers.

Fix: extend set_runtime_main() to accept base_url, api_key, and api_mode
keyword arguments; store them in new globals alongside the existing provider
and model; fall back to these globals in _resolve_auto() when the main_runtime
dict is empty. The call site in conversation_loop.py now passes all five
fields from the agent object.

Fixes #34777

* test(auxiliary): e2e routing assertions for custom-provider aux resolution

Adds two real-client tests on top of the salvaged #34783 fix:
- config-less custom:<name> endpoint routes via the carried live base_url
  (guards the #34777 symptom directly, not just the wiring)
- named custom:<name> WITH a config entry still resolves via the
  named-custom branch (regression guard against collapsing to bare custom)

* fix(tools): recognize email addresses as explicit targets in send_message

When using send_message with the email platform, valid email addresses
like user@example.com were not recognized as explicit targets by
_parse_target_ref(). This caused the function to return (None, None,
False), forcing the system into channel-name resolution which has no
way to resolve a raw email address, resulting in 'No home channel set
for email' errors.

Add _EMAIL_TARGET_RE pattern and email platform handler in
_parse_target_ref() so email addresses are treated as explicit targets
and routed directly without requiring a home target configuration.

* fix(tools): point email home-channel error at EMAIL_HOME_ADDRESS

The no-home-channel error for send_message derived the env var name
generically as <PLATFORM>_HOME_CHANNEL, producing EMAIL_HOME_CHANNEL for
the email platform. But gateway/config.py reads EMAIL_HOME_ADDRESS, so a
user following the error's guidance would set a variable that is never
consulted. Add a per-platform override map so the email hint names the
variable actually read; all other platforms keep the generic hint.

* perf(tui): stop slow/dead MCP servers from freezing TUI startup

The 'summoning hermes…' phase blocked on gateway.ready, which ran MCP
tool discovery inline. Any configured-but-unreachable MCP server burned
its full connect-retry backoff (1+2+4s ≈ 7s) before the composer
appeared — startup went from instant to ~7.5s of dead air for anyone
with a down stdio/http server in mcp_servers.

Move discovery into a background daemon thread so gateway.ready fires
immediately; tools register into the shared registry as servers connect,
and the agent isn't built until the first prompt. Measured spawn→ready:
~7500ms → ~115ms (dead twozero_td server in config).

Also drop rich.console + prompt_toolkit off banner.py's import path
(lazy-imported inside cprint/build_welcome_banner). tui_gateway.server
imports banner only to reach the lightweight prefetch_update_check
helper; the eager rich/pt imports added ~45ms before gateway.ready for
no benefit. tui_gateway.server import: ~115ms → ~69ms.

* feat(cli): add hermes prompt-size diagnostic (#35276)

Adds a 'hermes prompt-size' command that reports the fixed prompt budget
for a fresh session: system prompt total, skills index, memory, user
profile, prompt tiers, and tool-schema JSON bytes. Runs offline (dummy
credentials force the direct-construction path, no network call).

Lets users see which block dominates their per-call payload — the skills
index is often the largest single block when many skills are installed
(issue #34667). Zero model-tool footprint: it's a top-level CLI
subcommand, not an agent tool.

--platform <name> simulates a channel's platform hint; --json emits a
machine-readable breakdown.

Closes #34667

* fix(gateway): merge nested gateway.platforms configuration block

* fix(discord): bridge explicit allow_from configuration to env var mapping

* fix(gateway): run adapter config hooks for nested-only platform blocks

The plugin apply_yaml_config_fn dispatch loop only ran when a top-level
platform block (e.g. `discord:`) existed. Configs that defined a platform
only under `platforms.<name>` or `gateway.platforms.<name>` skipped the
hook, so `platforms.discord.extra.allow_from` never reached
DISCORD_ALLOWED_USERS. Fall back to those nested blocks when the top-level
one is absent.

Also map byquenox@gmail.com -> Que0x for the salvaged commits.

* fix(nous_account): add threading lock to prevent TOCTOU race on cache

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(file-tools): handle UTF-8 BOM in read_file / write_file / patch (#35278)

Some Windows editors prepend an invisible UTF-8 BOM (U+FEFF) to text
files. We had no awareness of it, so: read_file surfaced a phantom
U+FEFF as the first character; patch matches against the true first
line could miss; and a write/patch round-trip silently stripped the
marker, changing the file's byte signature.

Now:
- read_file / read_file_raw strip a single leading BOM so the model
  never sees it (only on the first chunk — the marker lives at byte 0).
- patch_replace strips the BOM before fuzzy-matching (so an exact
  first-line match works) and its post-write verification compares
  BOM-stripped content.
- write_file restores the BOM when the original file had one and the
  new content doesn't, mirroring the existing line-ending preservation
  (detect on disk via a cheap `head -c 3` probe or reuse pre_content,
  re-prepend across the edit). Guards against double-BOM.

Mid-content U+FEFF is left alone (it's data there, not a file marker).

Tests: TestBomHandling (real LocalEnvironment) — read-strips, raw-read
strips, write preserves, no-BOM-when-original-had-none, no-double-BOM,
patch round-trip preserves, patch matches first line through a BOM,
plus helper unit tests. 208 file-tool tests green.

* perf(read_file): compact line-number gutter — ~14% fewer tokens per read (#35368)

read_file's gutter used a fixed-width zero/space-padded prefix
("     1|content"). The padding is pure token overhead: measured with
cl100k on real Hermes source, the padded gutter costs ~48% more tokens
than bare content and ~16% more than a compact "<n>|content" gutter,
because the leading spaces tokenize into extra tokens on every line.

Switched the default to the compact "<n>|content" form. An A/B
(Sonnet 4.6 via OpenRouter, 2 passes, 4-task battery, every claim
verified against ground truth) showed:
  - padded  : 4/4 PASS both passes
  - compact : 4/4 PASS both passes  ← keeps line-referencing + patch
  - none    : 3/4 PASS both passes  ← dropping numbers entirely made
              the model hand-count lines and answer off-by-one (33 vs 34)

So we keep the line numbers (the model genuinely uses them to reference
lines) but drop the wasteful padding — capturing ~14% of the read-token
cost with zero measured accuracy change. Dropping numbers entirely
(the larger 33% saving) is rejected: it regresses line-referencing.

patch/fuzzy_match never consumed the gutter (they match old_string text
and compute char offsets internally), so editing is unaffected. No
downstream parser keys on the fixed-width columns. HERMES_READ_GUTTER=
padded restores the legacy format for anyone relying on alignment.

Tests: updated the 3 format assertions to the compact gutter; added an
env-override test for the legacy padded format. 209 file-tool tests green.

* fix(kanban): respect mobile safe areas in task detail drawer (#35378)

* fix(file-tools): handle UTF-8 BOM in read_file / write_file / patch

Some Windows editors prepend an invisible UTF-8 BOM (U+FEFF) to text
files. We had no awareness of it, so: read_file surfaced a phantom
U+FEFF as the first character; patch matches against the true first
line could miss; and a write/patch round-trip silently stripped the
marker, changing the file's byte signature.

Now:
- read_file / read_file_raw strip a single leading BOM so the model
  never sees it (only on the first chunk — the marker lives at byte 0).
- patch_replace strips the BOM before fuzzy-matching (so an exact
  first-line match works) and its post-write verification compares
  BOM-stripped content.
- write_file restores the BOM when the original file had one and the
  new content doesn't, mirroring the existing line-ending preservation
  (detect on disk via a cheap `head -c 3` probe or reuse pre_content,
  re-prepend across the edit). Guards against double-BOM.

Mid-content U+FEFF is left alone (it's data there, not a file marker).

Tests: TestBomHandling (real LocalEnvironment) — read-strips, raw-read
strips, write preserves, no-BOM-when-original-had-none, no-double-BOM,
patch round-trip preserves, patch matches first line through a BOM,
plus helper unit tests. 208 file-tool tests green.

* fix(kanban): respect mobile safe areas in task detail drawer

The task detail drawer is a body-level z-60 fixed overlay using
height:100vh starting at the viewport top. On mobile this puts the
drawer header behind the dashboard's fixed top bar (min-h-14, z-40)
and lets the bottom comment input sit under the browser's collapsing
nav bar.

- drawer: 100vh -> 100dvh (+ max-height:100dvh), 100vh kept as fallback
- head: padding-top honors env(safe-area-inset-top); mobile (<1024px,
  matching the lg breakpoint where the fixed bar shows) clears the
  3.5rem header
- comment-row + body: bottom padding extended with
  env(safe-area-inset-bottom) so the bottom-most element clears the
  mobile browser chrome

Mirrors the host shell idiom (100dvh + env(safe-area-inset-bottom) in
web/), and web/index.html already sets viewport-fit=cover so the insets
resolve. max()/calc() fallbacks leave desktop unchanged.

Closes #35324

* fix(gateway): recover model on post-interrupt turn; gate fallback status (#35381)

Empty model could reach the API on a recovery turn after stream_interrupt_abort,
failing HTTP 400 "No models provided" with no recovery — the session went
silent until the user manually re-sent (#35314).

- gateway/run.py: cache last-successfully-resolved model per session (+ a
  process-wide slot); when a fresh config read returns an empty model on a
  recovery turn, reuse the last-known-good instead of building model="".
- run_agent.py + agent/conversation_loop.py: only emit "trying fallback..."
  status when a fallback chain actually exists, so the UI stops announcing a
  fallback that will never run (also #17446).
- tests: empty-model recovery + _has_pending_fallback gate.

* fix(tools): wrap _run_tool cleanup in finally to prevent interrupt state leak

When _invoke_tool raises a BaseException (CancelledError, KeyboardInterrupt),
the cleanup code at the end of _run_tool was bypassed because it sat outside
the except block (which only catches Exception).  ThreadPoolExecutor recycles
thread IDs, so the leaked tid in _interrupted_threads poisons the next tool
scheduled on that thread — it instantly aborts with 'Interrupted'.

Move the discard + _set_interrupt(False) into a finally block so cleanup
runs regardless of how the worker exits.

Fixes #35309

* test(interrupt): assert no leaked tid instead of no-op block

Follow-up on the #35309 regression test: the trailing `with _lock: pass`
asserted nothing. Replace it with a concrete assertion that
_interrupted_threads is empty after the worker exits, directly verifying
the leak the fix prevents.

* fix(compression): drop conflicting 'resume Active Task' directive in summary prefix

SUMMARY_PREFIX previously contained two contradictory directives:

1. "treat it as background reference, NOT as active instructions"
   "Do NOT answer questions or fulfill requests mentioned in this summary"
   "Respond ONLY to the latest user message that appears AFTER this summary"

2. "Your current task is identified in the '## Active Task' section of the
    summary — resume exactly from there."

When the latest user message contradicted Active Task (e.g. 'stop the
i18n refactor', 'never mind, look at grafana instead'), models tended to
follow (2) anyway because 'resume exactly' is a strong, unambiguous
directive — leading to repeated re-surfacing of already-cancelled work
across turns, even after explicit 'stop'/'don't keep bringing that up'
messages from the user.

This change:
- Removes the conflicting 'resume exactly from Active Task' clause.
- Makes the precedence explicit: latest user message is the single source
  of truth; it WINS on conflict; cancelled Active Task / In Progress /
  Pending User Asks / Remaining Work must be discarded entirely (no
  'wrap up the old task first').
- Names canonical reverse signals (stop, undo, roll back, never mind,
  just verify, topic change) so the model recognizes them as cancellation
  triggers, not background context.
- Updates the summarizer template instruction so the LLM doesn't
  mechanically copy a cancelled task into Active Task on the next
  compaction (it's instructed to copy the reverse signal verbatim).
- Preserves: REFERENCE ONLY framing, MEMORY.md/USER.md authority, and
  the 'don't repeat work already reflected in session state' clause.

Adds tests/agent/test_summary_prefix_semantics.py to pin invariants so
the conflict can't regress.

Tested:
- All compaction tests pass: tests/agent/test_context_compressor.py,
  tests/agent/test_context_compressor_summary_continuity.py,
  tests/run_agent/test_413_compression.py,
  tests/run_agent/test_compression_persistence.py,
  tests/run_agent/test_compression_boundary_hook.py,
  tests/cli/test_manual_compress.py — 117/117 passing.
- Tested on macOS.

* fix(compressor): treat unanswered user questions as Active Task, not 'None'

The Active Task field in compression summaries is the single most important
field for task continuity across context boundaries. The previous template
described it narrowly as a 'task assignment' or 'request', which caused the
summary LLM to write 'None' whenever the user's most recent input was a
question, a decision request, or a discussion turn rather than an
imperative command. The assistant on the other side of the compaction then
treated the conversation as resolved and gave a generic recap instead of
answering the still-open question.

Expand the template guidance to cover:

  * explicit task assignments
  * questions awaiting an answer
  * decisions awaiting input (A vs B)
  * ongoing discussions where the assistant owes the next substantive reply

Reserve 'None' for the rare case where the last exchange was fully
resolved (e.g. user said 'thanks, that's all').

Also tighten the trailing CRITICAL instruction in the summary prompt so the
LLM cannot fall back to the old 'no imperative command → None' heuristic.

No behavioural code changes — template strings only. All 83 existing
compressor tests pass.

* fix(compressor): strip stale handoff prefix on resume; reconcile #26290+#32787 (#35344)

A handoff persisted under an older SUMMARY_PREFIX can be inherited into a
resumed lineage. _strip_summary_prefix only matched the current/legacy
literal, so on re-compaction the old 'resume exactly from Active Task'
directive stayed embedded in the body and kept hijacking replies to new,
unrelated user messages.

- Add _HISTORICAL_SUMMARY_PREFIXES (pre-#35344 prefix) and strip/recognize
  them in _strip_summary_prefix + _is_context_summary_content so resumed
  stale handoffs are re-normalized to the current latest-message-wins prefix.
- Reconcile the overlapping Active Task template edits from the salvaged
  #26290 (reverse-signal cancellation) and #32787 (capture open questions /
  decisions, don't write None too eagerly) — both intents kept.
- Regression coverage in tests/agent/test_resume_stale_active_task.py.
- AUTHOR_MAP entries for both salvaged contributors.

* fix(browser): recover from CDP DOM-node serialization crash in browser_console (#35385)

browser_console(expression="document.body") returned the cryptic CDP error
"Object reference chain is too long" instead of a usable result.

With returnByValue=true, Chrome deep-serializes the eval result; for a live
DOM Node/NodeList/Window that serialization overruns CDP's recursion guard
and fails the whole call with a protocol-level error (not a JS exception),
which _browser_eval surfaced raw.

- browser_supervisor.evaluate_runtime: on that specific error, retry once
  with returnByValue=false so Chrome returns the node's description string —
  the same graceful path already used for document.querySelector() results.
- browser_tool._browser_eval (CLI subprocess fallback): the subprocess can't
  retry, so convert the reference-chain error into actionable guidance
  (extract a primitive / use JSON.stringify) instead of leaking it raw.

No expression rewriting — normal evals (1+41 -> 42) are untouched.

* fix(cli): fail closed on empty oneshot responses

* fix(cli): surface oneshot agent exceptions to stderr with rc=1

Layer an exception guard on top of the empty-response fix so a crash
inside the agent (e.g. OSError from prompt_toolkit/Vt100 when stdout is a
non-TTY pipe, per #30623) is surfaced on the real stderr with rc=1 instead
of crashing past the redirect_stderr block. KeyboardInterrupt/SystemExit
are re-raised so Ctrl-C and explicit exits still propagate.

Also map briancl2 in scripts/release.py AUTHOR_MAP for the cherry-picked
empty-response commit.

Adapts the exception-guard approach from sweetcornna's PR #33818.

Co-authored-by: sweetcornna <96944678+ymylive@users.noreply.github.com>

* fix(dashboard): allow insecure WS peers on explicit non-loopback binds (#35386)

The merged 0.0.0.0/:: insecure-bind fix (#35141) did not cover binding
directly to a specific non-loopback address (e.g. a Tailscale/LAN IP via
--host 100.64.0.10 --insecure). In that mode the dashboard HTML loaded but
every WebSocket upgrade was rejected by the loopback-only peer guard, so
/chat connected then silently received no data.

Generalize _ws_client_is_allowed to lift the loopback-only peer gate for
any explicit non-loopback bound host, not just the 0.0.0.0/:: wildcard.
DNS-rebinding stays blocked: _ws_host_origin_is_allowed already requires
the Host header to exactly match the bound interface for explicit binds,
mirroring _is_accepted_host on the HTTP layer.

Co-authored-by: pxdsgnco <14163800+pxdsgnco@users.noreply.github.com>

* feat: add text debounce batching for WhatsApp and WeChat platforms

WhatsApp and WeChat (Weixin/iLink) both deliver messages individually
without any client-side batching, so rapid multi-message bursts (forwarded
batches, paste-splits, etc.) each trigger a separate agent invocation.

This wastes tokens (redundant system prompts / context for each fragment)
and degrades UX (the user receives reply fragments instead of a single
coherent response).

Both adapters now mirror the Telegram adapter's proven text-debounce
pattern:

- _text_batch_delay_seconds / _text_batch_split_delay_seconds
  (configurable via env vars)
- _pending_text_batches dict for per-session aggregation
- _enqueue_text_event() concatenates successive TEXT messages and
  resets the flush timer
- _flush_text_batch() dispatches after the quiet period expires

Configurable via env vars:
  HERMES_WHATSAPP_TEXT_BATCH_DELAY_SECONDS (default 5.0)
  HERMES_WHATSAPP_TEXT_BATCH_SPLIT_DELAY_SECONDS (default 10.0)
  HERMES_WEIXIN_TEXT_BATCH_DELAY_SECONDS (default 3.0)
  HERMES_WEIXIN_TEXT_BATCH_SPLIT_DELAY_SECONDS (default 5.0)

* fix(gateway): config.yaml path for WhatsApp/Weixin text-batch delays

Convert the salvaged text-debounce delays from HERMES_* env vars to
config.yaml (gateway.platforms.<name>.extra.text_batch_delay_seconds /
text_batch_split_delay_seconds), per the '.env is for secrets only'
policy. Adds a finite/non-negative guard so bad YAML values fall back to
the defaults instead of crashing asyncio.sleep().

- whatsapp.py / weixin.py: read delays via _coerce_float_extra(config.extra)
- update Weixin content-dedup regression test for the deferred dispatch path
- add text-debounce coverage (whatsapp + weixin): defaults, config override,
  bad-value fallback, env-var-ignored, burst-collapse, lone-message
- docs: WhatsApp + Weixin config keys

* fix(gateway): never auto-pause platforms on transient network/DNS failures (#35387)

The per-platform reconnect watcher auto-paused a platform after 10
consecutive reconnect failures, setting next_retry=inf and requiring a
manual /platform resume to recover. But both pause sites only ever fire
on *retryable* failures — non-retryable errors (bad auth) already drop
out of the retry queue earlier. So a transient DNS outage that spanned
the watcher's backoff window would silently park the bot forever, even
after connectivity returned.

The watcher's own docstring already promised 'retryable failures keep
retrying at the backoff cap indefinitely' — the code contradicted it.

Remove the auto-pause from both reconnect-failure branches. Retryable
failures now retry at the 5-min backoff cap forever and self-heal once
the network recovers. The circuit breaker (_pause_failed_platform /
_resume_paused_platform) stays for manual /platform pause|resume.

Fixes #35284.

* fix(gateway): support Windows absolute paths in MEDIA tag regex and extract_local_files (#34632)

The MEDIA_TAG_CLEANUP_RE and extract_local_files path regex both used
(?:~/|/) to anchor paths, which only matches Unix-style absolute and
home-relative paths. Two additional _TOOL_MEDIA_RE patterns in run.py
had the same limitation. Windows absolute paths (C:\Users\..., D:/...)
were silently ignored, causing MEDIA directive delivery to fail.

Add [A-Za-z]:[/\\] as a third anchor alternative in all four regex
locations (base.py x2, run.py x2). Also update path separators in
extract_local_files from / to [/\\] so it can traverse Windows
directory trees.

Revert accidental + quantifier in MEDIA_TAG_CLEANUP_RE lookahead
that changed match-one to match-one-or-more (unrelated to fix).

Fixes: #34632

* test: use raw docstring in test_run_tool_media_re to silence escape warning

* test: update extract_local_files Windows-path test for new matching behavior

test_windows_path_not_matched asserted the pre-fix POSIX-only behavior.
The Windows drive-letter support now intentionally matches these paths,
so replace it with parametrized positive cases plus a relative-path
negative guard, mirroring tests/gateway/test_platform_base.py.

* feat(kanban): file attachments on tasks (#35395)

Tasks can now carry file attachments (PDFs, images, source docs) that
workers read directly — closes the gap where source material had to be
pasted as a path into the task body.

- kanban_db: task_attachments table (additive), Attachment dataclass,
  add/list/get/delete accessors, attachments_root/task_attachments_dir
  path helpers (per-board, HERMES_KANBAN_ATTACHMENTS_ROOT override)
- build_worker_context: surfaces each attachment's absolute path so the
  worker (full file/terminal tool access) reads it via read_file/pdftotext
- dashboard API: POST/GET/DELETE attachment routes (multipart upload,
  25MB cap, traversal-safe filenames, root-containment check on download)
- dashboard UI: Attachments section in the task drawer — upload button,
  list with download, per-row remove
- docs + tests (13 cases: DB accessors, REST round-trip, traversal
  rejection, collision suffixing, worker-context surfacing)

Closes #35338

* perf(cli): stop eager MCP discovery from blocking agent-capable startup

* fix(file-tools): anchor relative-path resolution to absolute base; report resolved path (#35399)

Relative paths in write_file/patch could resolve against the agent PROCESS cwd
instead of the terminal's working directory. In a git-worktree session with a
stale TERMINAL_CWD='.' (a relative base), early edits silently landed in the
MAIN checkout, verified there, and reported success — while the agent inspected
the worktree and saw nothing, misreading it as the patch tool no-op'ing.

- _resolve_base_dir(): resolution base is now ALWAYS absolute. A relative
  TERMINAL_CWD is anchored to the process cwd once, deterministically, instead
  of being left to resolve()-time cwd. Live terminal cwd stays authoritative.
- write_file/patch pass the resolved absolute path to the shell FileOps layer
  so the tool layer and shell layer can't disagree about which file is edited.
- Responses now report the absolute resolved_path and files_modified, so a
  wrong-cwd mismatch is visible on the first call.
- _path_resolution_warning(): emits a _warning when a relative path resolves
  OUTSIDE the live terminal cwd (e.g. a worktree session writing into main).

Validation: 11 new unit tests + 43 live E2E assertions (worktree routing,
mid-session cd, V4A patches, divergence warning, absolute paths, consecutive
patches); 466 existing file/path/terminal tests green.

* fix(managed-gateway): keep tool availability scans off the Nous token-refresh path

* fix(cli): stop OSC 11 bg probe from trapping users in a stray editor (#35441)

Over SSH the OSC 11 background-color query round-trip routinely exceeds
the 100ms read budget, so _query_osc11_background() gives up and the late
reply lands after prompt_toolkit has grabbed the tty. prompt_toolkit then
injects the OSC payload as typed text and reads its BEL terminator
(\x07 = Ctrl+G) as a keystroke — Ctrl+G is the open-external-editor
binding, dropping the user into vi with garbage and no obvious way out.

- Skip the OSC 11 probe on remote sessions (SSH_CONNECTION/CLIENT/TTY);
  fall back to COLORFGBG / env hints / the dark default.
- Restore the tty with TCSAFLUSH instead of TCSANOW so any partial/late
  reply is scrubbed from the input buffer before pt reads it.

* perf(read_file): make compact gutter the only format; drop HERMES_READ_GUTTER (#35532)

The compact "<n>|content" gutter from #35368 is now the sole behavior.
Removes the HERMES_READ_GUTTER=padded escape hatch and its env lookup —
no legacy fixed-width path to maintain. Padding was pure token overhead
(~48% more tokens than bare content, ~16% more than compact) with no
measured accuracy gain in the original A/B.

- file_operations.py: drop env lookup + os import; gutter always f"{i}|{line}"
- tests: drop the padded env-override test; compact assertions retained

* fix(gateway): stop system tips from auto-uploading local files

* fix(gateway): denylist config.yaml for media delivery (belt-and-suspenders)

Defense-in-depth on top of the EphemeralReply gate: even if a config.yaml
path reaches response text via some other path, it can never be delivered
as a native attachment. Matches existing protection for .env, auth.json,
and credentials/.

Co-authored-by: JezzaHehn <jezzahehn@gmail.com>

* fix(install): refresh stale uv so installs actually get FTS5 Python (#35541)

The installer's ensure_fts5() handled a no-FTS5 Python by running
'uv python install --reinstall', but WHICH Python builds a uv can
install is baked into the uv binary's download manifest. A stale uv
(e.g. 'pip install uv==0.7.20', which predates python-build-standalone
#694) only knows about pre-FTS5 builds, so --reinstall just pulls the
same FTS5-less interpreter — a no-op for FTS5. Result: 'Could not obtain
an FTS5-capable Python' and a broken session search even on the
supported installer path.

ensure_fts5() now escalates uv itself: reinstall with current uv ->
'uv self update' + reinstall (stale standalone uv) -> install a fresh
standalone uv into a temp dir and reinstall with that (externally-managed
uv that can't self-update, the reported case). Pythons live in uv's
shared store, so the fresh uv's --reinstall overwrites the stale
interpreter in place and the installer's later 'uv python find' resolves
to the FTS5-capable build.

Verified against the reporter's exact repro (ubuntu:24.04 +
pip install uv==0.7.20): Python 3.11.13 (no FTS5) -> 3.11.15 (FTS5).

* fix(session): survive missing FTS5 runtimes

* fix(tui): swallow degraded mouse-burst noise so a stalled loop can't lock the composer (#35512)

* fix(tui): swallow degraded mouse-burst noise so a stalled loop can't lock the composer

When the Node event loop blocks during a heavy render/tool-call burst, stdin
stops being drained. Mode-1003 any-motion mouse reports pile up in the kernel
buffer, get partially read, and arrive as text with the `\x1b[<` prefix AND
coordinate digits chewed off across many partial reads. The existing fragment
recovery (SGR_MOUSE_FRAGMENT_RE) only handles clean `button;col;row[Mm]`
triples, so the degraded shards leak into the composer as typed text — the user
can no longer type or exit until the stall clears.

Captured leak (Windows Terminal, during tool calls):

  M6M35;220;56M6M35;218;56M169;48M;157;47M;44M20;43M79;40M78;40M0M7M35;49;41M
  48;41M;47;40M9;15;32M[I;31M5;211;26M35;211;25M7M;220;1MM0M09;25M24M23M3;22M
  M18M99;26M32MM38M63;44M47MM1;51M M4M54M

Add two recovery layers in parseTextWithSgrMouseFragments / the text-token path:

- MOUSE_BURST_NOISE_RE: whole-text fast path. If a text token is drawn only
  from the mouse-leak alphabet (`[ ] < ; I M m`, digits, spaces) AND carries
  the structural signature of mouse coordinates (>=3 M/m terminators, a digit,
  and a `;`), swallow it wholesale.
- MOUSE_BURST_RESIDUE_RE: swallows pure-noise residue in the gaps between and
  after recovered fragments, so a partially-recovered burst doesn't trail a
  chewed-up tail into the prompt.

All three constraints together preserve real prose: `Mmm MMM mmm yummy` has no
digit/`;`, `see 1;2;3M for details` has disqualifying letters, and
`1234;56;78M9;10;11M` has only two terminators — none are swallowed.

This is defense-in-depth: it stops the leak/lockout regardless of what blocks
the loop. The underlying event-loop stall during streaming is a separate,
still-open issue that needs live-turn instrumentation to root-cause.

* fix(tui): check mouse-burst noise before fragment recovery; drop test cast

Copilot review on #35512:

- MOUSE_BURST_NOISE_RE was only evaluated when parseTextWithSgrMouseFragments
  returned null. A noise blob that contains any intact `<b;c;r M` fragment makes
  fragment recovery return non-null, so the whole-text swallow never fired and
  the code emitted a pile of recovered mouse events instead of dropping the blob
  wholesale (contradicting the comment, and doing extra work mid-stall). Move the
  noise check ahead of fragment recovery so pure-noise tokens are dropped early.
  Add a regression test for a noise blob carrying intact fragments.

- Drop the unnecessary `(e as { isPasted?: boolean })` cast in the test;
  discriminated-union narrowing on `e.kind === 'key'` exposes isPasted directly.

* Potential fix for pull request finding

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

* Potential fix for pull request finding

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

* Potential fix for pull request finding

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

* Potential fix for pull request finding

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

---------

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

* fix(tui): clamp bogus terminal dimensions (WSL 131072x1) (#35657)

Some hosts (notably WSL) report a junk window size such as 131072 columns
by 1 row. Both the Ink fork and our components only guard against
0/null/undefined/NaN (stdout.columns || 80), so a positive-but-absurd
width sails through into createScreen(width*height), allocating tens to
hundreds of MB per frame and tripping the TUI memory monitor's hard exit.

Add clampStdoutDimensions(), installed in entry.tsx before ink.render: it
patches process.stdout.columns/rows with clamping getters (cols 1-2000,
rows 1-1000; out-of-range -> 80x24). One install point fixes the renderer,
its resize handler, and every component read. Live resizes still propagate
through the original descriptor, just clamped.

* fix(update): name new config options in migration prompt; skip prompt for pure version bumps (#35658)

The 'hermes update' config-migration prompt printed only counts ('1 new
config option available') then asked 'configure them now?' without ever
saying what the options were. Users said no because they couldn't tell what
they were agreeing to. For pure config-format version bumps (no new
env/config keys) it still asked the question, where saying yes just bumped
the version and looked like a no-op.

- List each new env var / config key by name + description before prompting
  (cap at 8, then '… and N more'). The data was already available; we just
  threw it away and printed a count.
- Pure version bump (no new options): apply the format migration
  non-interactively and print what happened, instead of asking a misleading
  yes/no.

Reported by ScottFive and Tt2021.

* fix: remove Discord mention redaction from secret scrubber

* test(redact): assert Discord mentions pass through unchanged

Rewrite TestDiscordMentions as negative assertions (mentions survive the
redactor) and clean up the orphaned comment + dangling whitespace left by
removing _DISCORD_MENTION_RE. Follow-up to the salvaged #32259 fix for #35611.

* feat(models): add deepseek-v4-flash, trim variants, group curated lists by maker (#35659)

* feat(models): add deepseek-v4-flash to OpenRouter + Nous curated lists

deepseek/deepseek-v4-flash was already in the native deepseek provider
catalog but missing from the curated OpenRouter and Nous Portal picker
lists. Added it to both and regenerated the model-catalog.json manifest
(drift guard requires same-PR regeneration).

* refactor(models): trim redundant variants, group curated lists by maker

Remove claude-opus-4.7/4.6, gpt-5.4-nano, gpt-5.3-codex,
gemini-3-pro-image-preview, gemini-3.1-flash-lite-preview, grok-4.20,
and the older gemini-3-pro-preview (Nous). Reorder both OPENROUTER_MODELS
and _PROVIDER_MODELS[nous] into contiguous per-maker blocks with comment
headers. Regenerated model-catalog.json (openrouter 27, nous 20).

* feat(models): add gemini-3-pro-preview to OpenRouter + Nous curated lists

Adds google/gemini-3-pro-preview to both curated pickers (new on
OpenRouter, restored on Nous). Regenerated model-catalog.json
(openrouter 28, nous 21).

* test(models): use claude-opus-4.8 in OpenRouter fetch fixtures

The two TestFetchOpenRouterModels tests mocked a live OpenRouter
response with claude-opus-4.6 and relied on it surviving the curated-list
filter. Since 4.6 was removed from OPENROUTER_MODELS, those models got
filtered out and the recommended tag shifted. Swap the fixture to
claude-opus-4.8 (still curated, still first in the Anthropic block).

* Block Hermes root config in media delivery

* fix(telegram): retry on httpx pool timeout instead of dropping the send (#35664)

When PTB's general httpx pool is exhausted, it converts httpx.PoolTimeout
into telegram.error.TimedOut whose message states the request was *not*
sent to Telegram. The send retry loop treated all non-connect TimedOut as
non-retryable, so a pool timeout raised immediately, skipped all 3 retry
attempts, and was returned as retryable=False -- silently dropping the
message (agent responses, cron reports, etc.).

A pool timeout means the request never left the process, making it the
safest case to retry. Add _looks_like_pool_timeout() and treat it like a
connect timeout in both the in-loop retry decision and the outer retryable
determination, so pool timeouts flow through the existing backoff loop and
stay retryable on exhaustion.

Reported-by: q3874758 (#35610)

* fix(security): neutralize file paths in mutation-verifier footer (#35584) (#35684)

The per-turn file-mutation verifier footer rendered failed-write paths as
bare absolute paths in the user-facing response. The gateway's
extract_local_files() scans response text for bare paths ending in a
deliverable extension (.yaml/.json/etc.), validates os.path.isfile(), and
auto-attaches matches as native uploads — so a denied write to
~/.hermes/config.yaml surfaced the path in the footer and got the
credential file silently uploaded to the messaging channel.

The gateway denylist (validate_media_delivery_path) already blocks the
config.yaml case after #35634. This is defense-in-depth at the source:
backtick-wrap every path the footer emits — both the bullet path and any
path echoed inside the tool's error preview (the protected-file denial
message embeds the path in single quotes, which do NOT block the
extractor regex). extract_local_files skips paths inside inline-code
spans, so wrapping defeats auto-attachment for ANY protected file while
keeping the path human-readable.

- run_agent.py: _format_file_mutation_failure_footer wraps bullet paths;
  new _neutralize_footer_paths backticks any remaining bare path (covers
  the preview echo). staticmethod -> classmethod (caller unaffected).
- tests: backtick-wrap assertion + end-to-end extract_local_files leak test.

* fix(gateway,cron): prevent agent restart loops via self-targeting gateway commands (#30719)

Three defenses against SIGTERM-respawn loops when agent schedules its
own gateway restart under launchd/systemd KeepAlive:

1. HERMES_IN_GATEWAY env var: gateway sets it at startup; stop/restart
   subcommands refuse to run when set (exit 1 with clear message).

2. Cron create payload filter: regex pre-flight rejects prompts/scripts
   containing hermes gateway restart/stop, launchctl kickstart/unload,
   systemctl restart/stop, and pkill patterns.

3. 30 new tests: pattern matching (14), cron block (5), gateway guard (4),
   safe command negatives (7).

* fix(gateway,cron): reuse existing _HERMES_GATEWAY marker; tighten cron regex

Follow-up to the salvaged #30728:
- Gateway already exports _HERMES_GATEWAY=1 at startup (gateway/run.py) and
  cli.py already keys off it. Drop the redundant new HERMES_IN_GATEWAY var;
  guard stop/restart on _HERMES_GATEWAY instead. One marker for one fact.
- Drop the greedy \bgateway.*restart alternation from the cron lifecycle
  filter — it false-positived on legit prompts that merely mention an
  unrelated gateway + a restart (API/payment gateway monitoring). The
  specific 'hermes gateway (restart|stop|start)' pattern already covers the
  real command.
- Rework the two negative guard tests to sentinel the first downstream call
  so they don't drive real signal delivery (tripped the live-system guard).
- Add false-positive regression cases to test_safe_commands.

* docs(toolsets): clarify all/* wildcard does not enable kanban (#35729)

The all/* wildcard expands to every registered toolset, but a handful of
tools have an additional check_fn gate on top of toolset membership and
are intentionally NOT turned on by all/* alone:

- Capability-gated tools (browser, computer_use, code_execution, Feishu,
  Home Assistant, cronjob) require their backend/credential prerequisite.
- The kanban toolset is workflow-gated and deliberately opt-in. Kanban
  tools mutate shared board state, so they stay off by default even under
  all/* — you must list 'kanban' by name (or be a dispatcher-spawned
  worker with HERMES_KANBAN_TASK set).

This was the expectations gap behind #35581 — the docs previously said
all/* expands to 'every registered toolset' without noting the carve-out.

Closes #35581.

* fix(voice): allow /voice over SSH when a sound server is reachable (#35719)

SSH sessions hard-failed voice mode on the presence of SSH_* env vars
alone, even when a PulseAudio/PipeWire server is running on the host and
audio works (ffplay/aplay/pw-play -> pulseaudio). Probe the default
sound-server sockets (PULSE_SERVER unix path, PULSE_RUNTIME_PATH/native,
$XDG_RUNTIME_DIR/{pulse/native,pipewire-0}) and actually connect() so a
stale socket doesn't count; downgrade the SSH branch to a notice when
audio is reachable. Mirrors the existing Docker/WSL forwarding handling.

Fixes #35622

* fix(vision): cap embedded image size before it wedges a session (#35732)

Resize vision tool-result images down to a 4 MB embed cap at load time,
not just at the 20 MB hard ceiling. A 5-20 MB image previously sailed
through the native fast path and got baked into conversation history,
where Anthropic's 5 MB per-image base64 limit rejected every subsequent
turn with a 400 — and because history is immutable, retries could never
clear it, permanently wedging the session.

Also harden the reactive shrink-recovery: it now returns False (don't
retry) when any oversized image part can't be brought under target, so
the single retry isn't burned re-sending a payload that will fail
identically. Previously it returned True after shrinking *any* part,
even when the actual oversized culprit survived.

* fix(streaming): stop duplicating tool-call args from cumulative-resend providers (#35718)

DeepSeek / Baidu Qianfan stream tool-call arguments in cumulative mode:
each chunk resends the full arguments-so-far instead of the new fragment.
The stream accumulator blindly concatenated arg deltas with +=, turning
that into '{...}{...}{...}', which failed json.loads and got nuked to '{}'
— a silently corrupted tool call (#35592). Worse on multi-param tools
(search_files, session_search, memory replace) because longer args take
more chunks, giving more resend opportunities.

- Per-slot cumulative latch in the stream accumulator: a delta that is a
  strict superset of the accumulated buffer marks the slot cumulative and
  replaces (not appends); exact duplicates are dropped only after latching.
  Incremental fragments are untouched (default += path).
- Backstop _collapse_repeated_json_arguments() in the repair pipeline
  collapses pure identical-resend buffers (K exact repeats of a valid-JSON
  unit) for providers that resend the complete object from chunk 1. Only
  reached after json.loads already failed, so compliant single objects are
  never touched.

Not a gateway or DeepSeek-model bug — any OpenAI-wire provider in
cumulative streaming mode is affected.

* feat(models): refresh model catalog hourly instead of daily (#35756)

Lower the model_catalog disk-cache TTL from 24h to 1h so freshly
published model-catalog.json deploys reach the picker within an hour
instead of up to a day. The picker now refetches on the next
`hermes model` / `/model` once the cache is older than 1h; younger
than 1h still serves the cache (no network hit), and network failures
still fall back to the stale copy.

- DEFAULT_TTL_HOURS 24 -> 1 (model_catalog.py)
- DEFAULT_CONFIG model_catalog.ttl_hours 24 -> 1, _config_version 24 -> 25
- migration v24->25 rewrites a stale ttl_hours:24 to 1, preserving any
  custom value the user set

E2E: verified >1h refetches / <1h skips, and migration rewrites 24->1
while preserving a custom 6.

* fix(feishu): cap _message_text_cache with LRU eviction to prevent unbounded growth

_message_text_cache was a plain dict with no size limit. Every unique
message_id whose text was fetched (for reply-context lookups) stayed in
memory permanently, causing unbounded growth in long-running deployments
with active group chats.

Replace with an OrderedDict and evict the least-recently-used entry
whenever the cache exceeds _FEISHU_MESSAGE_TEXT_CACHE_SIZE (512). Cache
hits call move_to_end() to refresh LRU order. Mirrors the identical
pattern already used by _pending_processing_reactions in the same class.

* fix(bluebubbles): cap _guid_cache with LRU eviction to prevent unbounded growth

The _guid_cache dict grows without bound as…