Tests version, config, agents, tools, prompts, skills, run, and help
commands. Skills list skipped (server-side 500), run skipped without
OPENAI_API_KEY in workspace secrets.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…esting

10 items: skills list 500, agent secrets not loading, missing secrets/config
CLI commands, missing get/describe for agents/tools/skills/prompts, and
undocumented workspace-scoped API routing via x-workspace-id header.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Cloud middleware only injects workspace model settings (default model,
provider secrets) when the A2A endpoint receives an agent UUID, not a name.
The CLI now fetches the agent config first to get the cloud UUID, then uses
that for streaming. Fixes `distri run` failing with "OPENAI_API_KEY not
configured" even when secrets and default model are set.

Updated bugs.md with root cause analysis and new missing feature (#3:
providers CLI command for configuring default model).

Integration tests: 14/14 passed, 1 skipped (skills list server bug).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

- Fix upsert_skill to try create directly when list_skills fails (server
  /skills endpoint returns 500), instead of hard-failing
- Add integration_summary.md with comprehensive test results: all CLI
  commands, 9 working tools, 6 failing tools, agent run tests, push tests
- Update bugs.md with 5 new server-side bugs found: secret injection only
  for system agent, search tool deserialization, container runtime missing,
  browsr_scrape schema mismatch, agent-internal tools in public list

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

- Use /v1/search instead of /search (migrated API)
- Build payload manually to avoid sending "limit": null (browsr expects
  usize, not Option<usize>)
- Use x-api-key header instead of Bearer for browsr-cloud auth

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

- Replace manual reqwest calls with BrowsrClient::from_env().search()
- No more duplicated URL/auth logic — all browsr tools use the client
- Switch to path deps for browsr-client/browsr-types (local testing)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Update tool schema and examples to only advertise formats that
browsr-cloud actually accepts: markdown, html, screenshot, structured,
agent. Removes summary, links, rawHtml, json, images, branding.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…sues

Full test matrix covering CLI, config, agents, tools (platform + browsr),
prompts, skills, and agent run. Includes environment setup instructions
and known failure documentation.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

- list_skills: handle both {"skills": [...]} (cloud) and [...] (oss) formats
- Move validate_provider_secrets before stream! macro so it runs with
  task-local user/workspace context from middleware (TenantSecretStore needs it)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Single source of truth for skill list types. Both distri-server (OSS)
and distri-cloud use the same SkillsListResponse wrapper format.
OSS server uses workspace_slug="local", full_name="local/{name}".

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Tests cloud format, defaults, and roundtrip serialization.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Tests were using old flat ModelSettings fields that moved into
ModelSettingsInner sub-struct. All 23 distri-types tests now pass.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

- Delete configuration/plugin_utils.rs (plugins removed)
- Inline is_namespaced_plugin_id/split_namespaced_plugin_id in orchestrator
- Fix ModelSettings tests to use inner struct
- All 17 distri-types tests pass

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

No more plugin system — simplify get_agent and agent_matches_simple_name
to direct name comparison only.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

15/15 integration tests pass, 0 skipped.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

- Fix BrowsrShellClient to use x-api-key header (not Bearer)
- Destroy session in background tokio::spawn — don't block tool response
- Make BrowsrShellClient Clone for sharing across tasks

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

browsr-shell already wraps with python3/node/bash based on language.
Remove wrap_code_for_execution and shell_escape — just send raw code.
Remove dead tests for removed functions.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

distri_execute_code now passes the user-provided language parameter
to the shell session instead of always auto-detecting. Auto-detection
still used as fallback when language not specified.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>