context-clearing
Skill0 starsChanged 3 months agoMITView on GitHub
Verify that a Loop-Dee-Loup PR, issue, or parent snapshot is a self-sufficient durable handoff before clearing, compacting, yielding, or replacing a session.
In this repository12 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- local-workerSkill
- model-checkSkill
- persona-makerSkill
- retroSkill
- script-makerSkill
- siftSkill
- skill-makerSkill
- spendSkill
- stage1-classifier-hardeningSkill
- telemetry-batterySkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 3 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 13
Verify it by direct read.
- Very specific: line 15
Run spend now if the completed session is worth measuring.
- Distinctive: line 18
Never clear first and hope the next session can reconstruct the work from comments or chat.
- Distinctive: line 39
Do not pull a full Codex audit report into a context-heavy orchestrator merely to learn CLEAN/NOT CLEAN.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Context clearing
Session context is a running cost. Clear or hand off when the needed state has been compressed into GitHub and the next step no longer benefits from the investigation or implementation history.
Safe order
- Open or update the PR, issue, or parent snapshot first.
- Verify it by direct read.
- Resolve any genuine founder decision before handing off.
- Run
spendnow if the completed session is worth measuring. - Clear, compact, end, or create a disposable successor. Do not keep both sessions working.
Never clear first and hope the next session can reconstruct the work from comments or chat.
Durable-record test
A successor with zero conversation memory must be able to recover:
- the intended outcome and non-goals;
- controlling authority and exact repository refs;
- what changed and why;
- acceptance and verification evidence;
- the current review stage and invocation count;
- accepted and rejected findings with root-cause disposition;
- unresolved blockers or founder decisions;
- the next authorized action.
A PR body also names the issue it closes. An audit issue names the exact merge commit and contains the complete read-only specification. A parent snapshot carries current truth rather than a transcript.
Review-cycle boundaries
The Stage 1 review request and CLEAN Stage 2 close are natural context boundaries. After requesting Stage 1, use PR event delivery where available and let a fresh session handle the response. For a Stage 2 issue audit, delegate the trigger, bounded wait, and compact verdict extraction to one disposable context when possible; issue responses may not have PR event delivery.
Do not pull a full Codex audit report into a context-heavy orchestrator merely to learn CLEAN/NOT CLEAN. Return only the verdict, severity counts, and finding text needed to act.
Autonomous handoff
When the environment can create a successor session and no human is present to clear manually, create exactly one successor only after the durable-record test passes. Its prompt names the repository, exact PR or issue URL, current stage, and governing files. It must read the durable record itself; do not paste a conversation summary. End the old session immediately after the handoff.
This is not a way to defer an unresolved decision, avoid required verification, or run concurrent duplicate controllers.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.