session-distill

name	session-distill
description	Distill Hermes Agent SQLite sessions into review packets, local session notes, and reusable memory candidates. Use when the user asks to整理/提炼 Hermes 对话, process `%LOCALAPPDATA%\hermes\state.db`, generate session packets, review packet coverage, or mark Hermes sessions as distilled.
argument-hint	[status | run --next N | deep-distill | mark <id> distilled]
user-invocable	true

Hermes Session Distiller

This is the Hermes-native session-distill skill. It reads Hermes state.db (SQLite), not Claude Code / Codex / Cursor session files.

Inputs And Outputs

• Input sessions:
  • %LOCALAPPDATA%\hermes\state.db (override: HERMES_STATE_DB)
  • optional HERMES_HOME when not under default LocalAppData path
• Workspace:
  • %LOCALAPPDATA%\hermes\session-distill\manifest.json
  • %LOCALAPPDATA%\hermes\session-distill\packets\<session-id>.md
  • %LOCALAPPDATA%\hermes\session-distill\distilled\sessions\<session-id>.md
  • %LOCALAPPDATA%\hermes\session-distill\knowledge-base.md
  • optional %LOCALAPPDATA%\hermes\session-distill\memory-drafts\<session-id>.json

Default Workflow (Deep Distill)

Follow references/deep-distill-workflow.md (shared across platforms):

1. python bin/deep-distill-run.py --batch-size 3 — bundle + extract claims → distilled/answer-packets/
2. answer-me: verify each Q with Read/Grep/git/Shell; only ANSWERED promotes
3. Session note under distilled/sessions/<session-id>.md
4. distilled/check-work/batch-*-report.md with promoted / not-promoted reasons
5. mark <session-id> distilled after check-work PASS

Legacy one-shot flow (run --next 1 without answer-packets) is for packet preview only — do not auto-promote to KB.

Packet Preview (optional)

1. Run python bin/hermes-session-distill.py run --next 1.
2. Read the generated packet and inspect Packet Audit first.
3. If Coverage: partial, inspect the raw SQLite conversation span before promoting any conclusion.

mark distilled requires a session note and promotion decision. Partial packets must document raw DB review.

Commands

$env:HERMES_DISTILL_DIR = "$env:LOCALAPPDATA\hermes\session-distill"

python $env:LOCALAPPDATA\hermes\skills\session-distill\bin\deep-distill-run.py --offset 0 --batch-size 3
python $env:LOCALAPPDATA\hermes\skills\session-distill\bin\hermes-session-distill.py status
python $env:LOCALAPPDATA\hermes\skills\session-distill\bin\hermes-session-distill.py list --size 100
python $env:LOCALAPPDATA\hermes\skills\session-distill\bin\hermes-session-distill.py run --next 1
python $env:LOCALAPPDATA\hermes\skills\session-distill\bin\hermes-session-distill.py mark <session-id> distilled
python $env:LOCALAPPDATA\hermes\skills\session-distill\bin\hermes-session-distill.py self-test

References

• references/deep-distill-workflow.md: canonical Deep Distill pipeline (all platforms).
• references/distillation-rules.md: promotion and filtering rules.
• references/output-layout.md: workspace layout and status meanings.

Use other platform distillers only for their respective session roots.