Summary

Implements #719 — when resuming a session via --continue or --resume, displays a compact recap of previous conversation messages in a Rich panel before the input prompt.

Before: User sees only a one-line status and has zero context about what was discussed.
After: A styled "Previous Conversation" panel shows user messages, assistant responses, and tool call summaries immediately upon resume.

What Changed

cli.py (3 files, +700 lines)

New methods on HermesCLI:

• _preload_resumed_session() — Loads session history from SQLite early (called from run() before the banner), so the user sees history immediately. Prints the one-liner status message. Returns True if history was loaded. _init_agent() now skips the DB round-trip when history is already populated.
• _display_resumed_history() — Renders a compact conversation recap using a Rich Panel:
  • User messages shown with gold bullet ●, truncated at 300 chars
  • Assistant responses shown with green diamond ◆, truncated at 200 chars / 3 lines
  • Tool calls collapsed to count + tool names (e.g. [3 tool calls: terminal, web_search])
  • System messages and tool results hidden
  • <REASONING_SCRATCHPAD> blocks stripped from assistant text
  • Pure-reasoning messages (no visible output after stripping) skipped entirely
  • Capped at last 10 exchanges with ... N earlier messages ... indicator
  • Dim/muted styling so recap is visually distinct from the active conversation

Integration in run():

• After show_banner(), if self._resumed: calls _preload_resumed_session() then _display_resumed_history()
• History is displayed before the user types anything

Instance variable:

• self.resume_display — captured from config at init time (like self.compact)

hermes_cli/config.py

• Added display.resume_display to DEFAULT_CONFIG with default "full"

Config option

display:
  resume_display: full    # full | minimal

• full (default) — show previous messages panel
• minimal — current behavior (just the one-liner)

Test Plan

27 new tests in tests/test_resume_display.py:

• Display tests (17): user/assistant rendering, system/tool message hiding, tool call summaries, message truncation, multiline truncation, large history truncation indicator, multimodal content, empty history, minimal config, panel title, reasoning stripping, pure-reasoning skip, mixed text+tool_calls
• Preload tests (7): not-resumed guard, no-DB guard, session-not-found, no-messages, successful load, DB reopening, singular grammar
• Integration (1): _init_agent() skips DB when preloaded
• Config (2): DEFAULT_CONFIG and cli defaults include resume_display

$ python -m pytest tests/test_resume_display.py -v
27 passed in 0.67s

$ python -m pytest tests/ -q
2371 passed, 5 skipped, 23 deselected in 113s

Interactive verification: Tested with python cli.py --resume <session_id> against real sessions with tool calls, reasoning content, and context summaries. Display renders cleanly.

Example Output

╭─────────────────────────── Previous Conversation ────────────────────────────╮
│   ● You: Read all 4 GitHub issues (#501, #502, #503, #504) from              │
│ NousResearch/hermes-agent in full. For each issue, extract: (1) the exact    │
│ scope/phases...                                                              │
│   ◆ Hermes: [4 tool calls: terminal]                                         │
│   ◆ Hermes: All 4 issues read successfully. Here is the structured analysis: │
│             ================================================================ │
│ ======== ...                                                                 │
╰──────────────────────────────────────────────────────────────────────────────╯

Closes #719