agentleFS
igovetigovet/cortex-orch-acp/skills/knowledge-harvest-scope/SKILL.md

knowledge-harvest-scope

Skill1 starsChanged 59 days agoMITView on GitHub

Scope explicit harvest and harvest-refresh requests from retained evidence, without coordinator repository census.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    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
  1. Distinctive:

    Only explicit /harvest or /harvest-refresh authorizes this work.

    line 8
  2. Distinctive:

    After writing, compare the current source inventory with the coverage matrix, feature pages, links and checks.

    line 45
  3. Very specific:

    Do not use a task-level documentation not required outcome for a completed harvest, because documentation is the requested artifact.

    line 77

≈ “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

Harvest scope assessment

Only explicit /harvest or /harvest-refresh authorizes this work; optional text narrows scope and no previous task is required. Ordinary product work does not authorize a census or broad documentation rewrite. Harvest builds a complete source-backed map; refresh rebuilds the inventory against current source, corrects stale knowledge and finishes with a post-write comparison.

For harvests only, repository reads, source analysis, edits, commands and completeness checks are worker-owned. Read report previews, the pipeline and bounded evidence pages needed for a scope decision, not project indexes or runtime implementation. Missing knowledge indexes justify bounded census work, not coordinator source inspection. This restriction does not apply to ordinary orientation, direct answers or evidence reconciliation. Require harvest workers to read knowledge-harvest before census work; it owns the canonical output layout, complete inventory and coverage contract. Workers confirm documentation in current source and report unresolved gaps, not guessed coverage percentages.

Scale the assignment to the repository

Choose the fewest assignments that can establish complete, source-backed coverage. A small repository with a bounded source tree, one runtime, and few feature-bearing surfaces can use one technical-writer worker for the whole harvest or refresh. Assign that worker to enumerate the current tree, inspect source, tests and configuration, map every in-scope surface, write or update the canonical documents, run relevant deterministic checks, and report exact gaps and outcomes. This is one cohesive assignment, not a sequence of planner, explorer, architect, writer, reviewer and verifier handoffs. Its self-checks are evidence of what ran, not independent review.

For a broad or coupled repository, partition only when the source tree or ownership makes one assignment impractical. Delegate bounded explorer domains where independent discovery reduces uncertainty; use planner for a real scope or dependency problem and architect for a consequential boundary decision. Give exactly one writer ownership of shared docs/project/ and docs/features/index.md; other writers may own disjoint feature pages. A worker needing earlier evidence receives only the relevant finalized reports.

After writing, compare the current source inventory with the coverage matrix, feature pages, links and checks. For each correction to a documented behavior, default, or output (including one found against unchanged source), inventory every relevant occurrence across the target repository's documentation, including otherwise unchanged pages, commands, flags, example inputs and expected output. Confirm each claim in current source and correct stale examples. Inventory every affected executable example across those pages; run safe examples against the current source and compare exact exit status and stdout with their documented expectations. Mark examples that cannot safely run or whose output cannot be verified as unresolved, naming the reason. The writing worker may perform this comparison and repair bounded gaps. The runtime supplies an independent reviewer for these changed documentation claims. The reviewer checks current source and post-write documentation, including the complete relevant example inventory and observed results, without relying on the writer's claim that a search was clean. Propose additional specialist investigation for a concrete unresolved risk, such as ambiguous feature boundaries, conflicting source evidence, unexplained surfaces, consequential safety behavior, or checks the writer cannot run or interpret. Do not repeat a check on an unchanged revision without a reason. A small refresh may use one writing assignment; independent review and command verification remain mandatory runtime stages. Report unresolved gaps honestly. Models propose adaptations; the runtime owns scheduling, review and completion.

Relationship to the final documentation stage

A harvest or harvest-refresh is documentation-impacting by definition. Its documentation delegations serve as the dedicated documentation-sync worker or workers for docs/project/, docs/features/, and any affected public documentation. After the final writing or corrective delegation, the coordinator assesses the worker's source comparison and checks and submits a synthesis grounded in retained evidence. The runtime requires independent review and automatic verification before completion. It may follow bounded report cursors for facts needed by that decision, while documentation indexes and project files remain worker-owned. Missing specialist interpretation requires a bounded worker check.

Do not use a task-level documentation not required outcome for a completed harvest, because documentation is the requested artifact. Public documents that the worker reports as unaffected should remain unchanged rather than receiving meaningless edits.

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.