agentleFS
igovetigovet/cortex-orch-acp/skills/context-compaction/SKILL.md

context-compaction

Skill1 starsChanged 59 days agoMITView on GitHub

Restore coordinator and specialist context after compaction or restart.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

9 rules. About 0 are common, 4 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use this skill after context loss.

    line 8
  2. Very specific:

    Call task_get_context to recover the current task, requirements, plan revision, results and unresolved decisions in the same session.

    line 16
  3. Very specific:

    Read the current pipeline with read_report, following bounded cursors only for the original goal, decisions and evidence the next action …

    line 18
  4. Very specific:

    Use list_sources and needed source reads only for exact user requirements and attachment references the summary lacks.

    line 20
  5. Distinctive:

    Follow the runtime's per-turn language directive (English or Russian), not the language of a recovery summary.

    line 24
  6. Very specific:

    Read read_capsule first when the summary lacks the exact assignment, role instructions, requirements, declared work area, retained checks, …

    line 36
  7. Very specific:

    Read the assignment's relevant immutable reports and clarifications through bounded pages.

    line 39
  8. Distinctive:

    If your role writes, inspect current target state before editing and preserve unrelated changes.

    line 41
  9. Distinctive:

    Use the runtime's advertised routes within their live contracts for missing input or additional project paths.

    line 48

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Context recovery

Use this skill after context loss. A recovery summary points into durable state; it does not replace the original request, assignment or evidence. Recover only what the summary lacks for the next action, in the order below, and stop once you have it. Reload the next needed live tool declaration before calling it; remembered arguments are not authoritative.

Coordinator

  1. Call task_get_context to recover the current task, requirements, plan revision, results and unresolved decisions in the same session.
  2. Read the current pipeline with read_report, following bounded cursors only for the original goal, decisions and evidence the next action needs.
  3. Use list_sources and needed source reads only for exact user requirements and attachment references the summary lacks. Read relevant immutable reports by exact reference.

Follow the runtime's per-turn language directive (English or Russian), not the language of a recovery summary. Worker messages and briefs stay in English. The coordinator may read bounded project files or code-index results for a direct answer, orientation or evidence reconciliation, consistent with the orchestrator. Harvest-specific source discovery remains worker-owned. Project edits and checks belong to workers; the runtime owns scheduling, review and verification. Do not recreate workers, manage engine handles, replay uncertain external effects or infer completion from chat. Report missing or inconsistent evidence; the runtime reconciles durable identities.

Worker

  1. Read read_capsule first when the summary lacks the exact assignment, role instructions, requirements, declared work area, retained checks, report references or drafts.
  2. Read the assignment's relevant immutable reports and clarifications through bounded pages. Use list_reports only when a needed reference is missing.
  3. If your role writes, inspect current target state before editing and preserve unrelated changes. Reuse retained command receipts rather than rerunning unchanged checks. Consultants recover only their packet and retained reports; do not inspect project source or perform checks.

Resume English worker communication; user-facing structured questions follow the runtime's per-turn language directive and advertised tool contract. Recovery does not grant new scope or access. Use the runtime's advertised routes within their live contracts for missing input or additional project paths; do not ask the user to repair tool access. Report unavailable sources or attachments explicitly; do not infer their contents or create a replacement task. Retain owned draft identities and acknowledged publications.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.