documentation-impact
Skill9 starsChanged 37 days agoMITView on GitHub
Check documentation impact before completing a change or declaring it ready to ship. Use when a change may affect user, operator, or developer instructions, when asked whether docs need updates, or for a documentation pass. Supports author and read-only review modes without requiring a PR or spec.
2 other projects use a version of this.
Adopted by
Most recently taken up by
- sodejm/copilot-operation-plugin-for-securitytheir own version, first seen Oct 23, 2026
- sodejm/portable-agent-repository-kittheir own version, first seen Sep 1, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- sodejm/copilot-operation-plugin-for-security★ 2same name
- sodejm/portable-agent-repository-kit★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository64 other files it gives its agents.
- architectSkill
- arenaSkill
- automate-meSkill
- benchmark-checklistSkill
- blast-radiusSkill
- broSkill
- correctSkill
- create-formal-verification-skillSkill
- create-verification-skillSkill
- deslopSkill
- figure-it-outSkill
- howSkill
- interrogateSkill
- in-your-own-wordsSkill
- ketchupSkill
- ketchup-videoSkill
- maintain-formal-verification-skillSkill
- maintain-verification-skillSkill
- make-bot-uiSkill
- no-commentsSkill
- poteto-helpSkill
- poteto-modeSkill
- poteto-tddSkill
- poteto-teachSkill
- principle-attack-the-premiseSkill
- principle-boundary-disciplineSkill
- principle-build-the-leverSkill
- principle-encode-lessons-in-structureSkill
- principle-exhaust-the-design-spaceSkill
- principle-experience-firstSkill
- principle-explain-the-numberSkill
- principle-fix-root-causesSkill
- principle-foundational-thinkingSkill
- principle-guard-the-context-windowSkill
- principle-laziness-protocolSkill
- principle-make-operations-idempotentSkill
- principle-migrate-callers-then-delete-legacy-apisSkill
- principle-minimize-reader-loadSkill
- principle-model-the-domainSkill
- principle-never-block-on-the-humanSkill
- principle-outcome-oriented-executionSkill
- principle-prove-it-worksSkill
- principle-redesign-from-first-principlesSkill
- principle-separate-before-serializing-shared-stateSkill
- principle-sequence-verifiable-unitsSkill
- principle-subtract-before-you-addSkill
- principle-test-behavior-not-implementationSkill
- principle-type-system-disciplineSkill
- pr-recapSkill
- pstack-harnessSkill
- recallSkill
- reflectSkill
- runtime-probesSkill
- setup-pstackSkill
- sharpenSkill
- show-me-your-workSkill
- spec-probesSkill
- statechartSkill
- swarmSkill
- technical-writingSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 4 are distinctive, and 2 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.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 12
Use author when the task authorizes documentation updates.
- Distinctive: line 19
Use the accepted request, issue, or existing work brief.
- Distinctive: line 34
Read the repository's instructions and discover where its maintained docs and examples live.
- Distinctive: line 36
Consider these document types where the change affects them:
- Distinctive: line 46
Search for old names, old values, and contradictory examples beyond files already changed.
- Very specific: line 77
Run a full review when a repair changes behavior or contract claims, or when that effect is uncertain.
≈ “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
Documentation impact
Match the documentation to the behavior people will use. Inspect the actual change and its callers before deciding which docs need work. A supported no-change conclusion is a complete result.
Choose the mode and scope
Use author when the task authorizes documentation updates. Use review for an audit, independent review, status request, or other read-only task. Honor the caller's mode. If update authority is unclear, inspect and report in review mode.
- In
authormode, update affected documentation within the authorized change. Leave unrelated edits alone. - In
reviewmode, inspect without editing the artifact you judge. Return findings to its author. Do not fix a document and certify your own fix.
The calling workflow owns delegation, model selection, continuation, and human gates. Do not spawn another coordinator or start a publication workflow here.
Use the accepted request, issue, or existing work brief. Do not require a new spec, ticket, docs tree, or work ledger.
For a repair recheck, use the retained coverage and repair rules below to scope the reader mapping and inspection steps. Reopen the original full-review diff only when those rules require a full review.
Bind the pass to what you actually examine:
- For a full review of committed work, resolve and record the fixed base and head. Inspect the base-to-head diff and the surrounding implementation and documentation at those revisions.
- For a repair recheck, record the original full-review span as lineage. Use the previously reviewed artifact as the comparison base and inspect its delta to the resulting artifact.
- For local work, also inspect staged and unstaged changes and relevant untracked files. Record a content fingerprint or retained snapshot of those inputs. A clean-looking HEAD or a filename list does not identify dirty file contents.
- Outside Git, record an explicit content snapshot of the scoped implementation and docs. If a relevant input is unavailable, name the gap instead of claiming coverage.
Record the examined paths and either the fingerprint method or the retained snapshot location so another agent can reproduce the comparison. Recheck the inputs before returning the result. If concurrent edits changed them, refresh the affected checks or report that the result no longer describes the current artifact.
Map the change to readers
Read the repository's instructions and discover where its maintained docs and examples live. Trace changed behavior to the people who use, run, upgrade, or develop the project. Include public workflow changes in skills and configuration, even when no application code changed.
Consider these document types where the change affects them:
| Changed behavior | Candidate documentation |
| Installation, onboarding, or basic usage | README, quickstarts, tutorials |
| Commands, APIs, flags, defaults, configuration, or environment variables | CLI and API reference, configuration guides, executable examples |
| Compatibility, storage, or upgrade behavior | Migration and upgrade instructions, deprecation notices |
| Release-visible behavior | Release notes or changelog, if the repository maintains them |
| Module boundaries, data flow, or operational behavior | Architecture explanations, diagrams, runbooks, developer guides |
Search for old names, old values, and contradictory examples beyond files already changed. Read the matched claims in context. A recent timestamp, a path match, generated prose, or a rendered diagram cannot establish factual accuracy.
For each affected reader task, identify the maintained document or explain why no document needs a change. Existing docs may already be correct. Internal refactoring may leave every documented contract unchanged. Do not create filler docs to make the pass appear productive.
Update or inspect
In author mode, make the smallest corrections needed for the changed behavior. Follow the repository's documentation layout and generation commands. Update the source of generated docs instead of hand-editing generated output.
Call the Skill tool with technical-writing on documentation you write or review. Then call the Skill tool with unslop on it. Keep examples tied to real symbols, paths, flags, defaults, and supported versions. Example paths must resolve on a reader's machine. For machine-specific paths, use an explicit placeholder and show how to discover or supply the real value. A temporary path is valid when a preceding step creates it. A writer's unexplained scratch path is not a usable example. If an upgrade requires user action, state that action and the old-to-new behavior.
Trace documented effects through the implementation. Accepting or reporting a setting does not prove that the corresponding operation runs. If intended behavior and observed behavior disagree, report the mismatch as an implementation finding instead of documenting a defect as supported behavior.
In review mode, derive the affected reader tasks from the implementation yourself. Check the author's coverage decisions against the source, including claims that nothing changed. Treat the author's receipt as evidence to inspect, not as the verdict.
Use the repository's existing diagram format when a diagram explains an affected relationship better than prose. Check each changed actor, edge, direction, and boundary against the implementation. Optional visual tools can help present findings if available. Otherwise use Markdown or the existing diagram source. Do not install a visual dependency, require branding input, or wait for annotation feedback to complete this pass.
Decide whether author mode requires review
An author pass returns exactly one result:
independent review requiredwhen the change edits maintained documentation, changes a public workflow or a documented contract such as a default or flag, or leaves a documentation gap unresolved.independent review not requiredonly when evidence shows that none of these holds.
Record the evidence that supports the author result with the examined artifact. A completion or shipping workflow can start an independent review only when the author result is independent review required.
When independent review is required, a different agent reviews the resulting artifact without editing it. The first review covers the full accepted scope and reports every finding in that one pass. Return pass, needs changes, or unverified. Report a wording suggestion that leaves no claim wrong or missing as a note. A note does not block pass.
After a repair, retain the previous review, its examined artifact, findings, and valid coverage. The repairer records the repair delta and searches for other copies of each corrected claim. Retain the search terms, searched paths, and matches read in context, including a search with no matches.
For wording-only repairs, give the next reviewer that evidence and the resulting artifact identity. Recheck the repair delta and the repairer's search, with the context needed to judge them. Carry forward coverage of unchanged prose instead of reviewing it again. Resolve every outstanding finding before returning pass on the resulting artifact.
Run a full review when a repair changes behavior or contract claims, or when that effect is uncertain. Missing prior coverage or repair-search evidence leaves the recheck unverified until supplied or replaced by a full review. Do not infer a pass from a round count.
Allow one repair round per review. The recheck after that round, scoped or full, returns needs changes when a finding remains unresolved or it establishes an omission or incorrect claim. It reports wording suggestions as notes and returns pass when only notes remain. When the recheck does not return pass, return its open findings to the caller as pending instead of starting another round.
Bind every recheck to the resulting revision or content snapshot and the accepted scope. Verify that the carried coverage still matches the unchanged inputs and relevant surrounding context. Concurrent edits or changed context invalidate the affected coverage. Refresh those checks or run a full review before returning pass. A matching stable patch-id after a rebase does not establish unchanged documentation context.
Verify and return the result
Run the relevant repository documentation checks and exercise changed examples against the actual implementation when safe within the task's authority. Call the Skill tool with verify-commands on commands used as completion evidence. In review mode, run checks without modifying the reviewed files, using a disposable copy if the check writes output.
Check factual meaning as well as syntax and rendering. An example that runs with the wrong default or a diagram that invents an approval step is still wrong. Report unsafe or unavailable checks as gaps. Never run a deployment, publish docs, or change external data just to verify an example.
Return a compact result in the conversation or the existing work record:
- The mode, scope, and examined revision or snapshot, including relevant local changes.
- The affected reader tasks and document paths, with updates made or evidence for leaving them unchanged.
- The checks actually run, their outcomes, and any remaining findings or verification gaps.
- In author mode, the author result and its evidence. In review mode, return
pass,needs changes, orunverified. Useneeds changesfor established omissions or incorrect claims. Useunverifiedwhen missing evidence prevents a decision.
An author result is not an independent review verdict.
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.

