daseinlabs/parsec/trim
Claude Code and agent skill by daseinlabs/parsec
- community
Checking that the SKILL.md file is still in the repository.
Stage a det+dir compaction of this session — a deterministic needed-set trim of the transcript plus a standing-directives block, injected automatically after /clear. Use when context is nearly full and the user wants a better /compact.
Install
npx vskill@latest install daseinlabs/parsec/trimSKILL.md
Follow these steps in order. Every number you report from step 1 is an ESTIMATE (chars/4) — always say "estimated", never present it as measured.
If the parsec MCP tools trim_stage and trim_finalize are available, use them instead of the shell commands below: trim_stage (optionally with level) replaces step 1 and trim_finalize (directives as the argument) replaces step 3 — no heredoc needed. Everything else is unchanged.
- Run
"${CLAUDE_PLUGIN_ROOT}/bin/parsec" trim --json.
- If the user asked for a specific trim aggressiveness, add
--level N
(1 = low trimming / keep more, 5 = very high). Omit it otherwise — the default (3) is the only measured configuration; the other levels shift what counts as "used later" and are estimates all the way down.
- Exit code 2 means the session is too short to trim (or no transcript was
found) — relay the binary's message plainly and stop.
- Otherwise present the stats conversationally: kept X of Y chunks,
roughly A of B estimated tokens. This is the deterministic trim — the parts of this session that were actually re-read, edited, or used later.
- Now write the STANDING DIRECTIVES block from your own in-context history.
You are the compaction assistant here; the history is already in your context. Your job is NOT to summarize the work. Extract ONLY the STANDING DIRECTIVES: durable rulings, policies, quantitative floors, model/tool choices, scope decisions, and the north-star — the things the user established that must keep governing behavior AFTER compaction. Exclude one-off task steps and transient state. Output a terse pinned-constraints list, each line an imperative the agent must keep obeying. If a directive was stated with force or repeated, keep its force. Keep the list under roughly 3,000 tokens.
- Stage the directives (heredoc keeps the artifact on disk complete before
anything else happens):
`` "${CLAUDE_PLUGIN_ROOT}/bin/parsec" trim --finalize <<'PARSEC_DIRECTIVES' <the directives list> PARSEC_DIRECTIVES ``
- When the binary prints
ready, tell the user: run/clearnow — the next
session in this project starts with the trimmed context injected automatically. Mention it expires unused after 30 minutes, and that it fires on /clear (or a fresh start), not on resume.
- If step 3 fails for any reason, say so plainly and tell the user the
deterministic trim alone is still staged and will be injected without the directives block after /clear (that is the designed fallback, not an error state).
Never estimate savings the ledger does not show, and never run /clear yourself — the user decides when the session ends.
Security
- Security scanPassed, 95 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingpackages/plugin/skills/trim/SKILL.md
Popularity
- Repository stars
- 82
- Repository forks
- 39
- 7-day trend score
- 22
Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/daseinlabs/parsec/trim)