autoevo-review
Skill5 starsChanged 35 days agoMITView on GitHub
Triage actionable Autoevo findings; empty cycles remain machine receipts.
In this repository38 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- atelierSkill
- autoevo-reviewSkill
- civSkill
- curateSkill
- daily-reflectionSkill
- decisionSkill
- digestSkill
- dineSkill
- hiSkill
- introspectSkill
- lintSkill
- prmSkill
- pushSkill
- readSkill
- reflectSkill
- reviewSkill
- triageSkill
- weeklySkill
- captureSkill
- readingSkill
- civSkill
- curateSkill
- daily-reflectionSkill
- decisionSkill
- digestSkill
- dineSkill
- hiSkill
- introspectSkill
- lintSkill
- prmSkill
- pushSkill
- readSkill
- reflectSkill
- reviewSkill
- triageSkill
- weeklySkill
The file
We split this skill into the rules it gives an agent
9 rules. About 0 are common, 3 are distinctive, and 6 are 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.
6
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 8
Follow protocols/autoevo.md.
- Very specific: line 11
Never edit canonical nightly JSON or its derived reports during review.
- Very specific: line 22
Before presentation, stage autoevo_pending.py auto-dismiss --today <date> on private state for entries at three surfaces or over the …
- Distinctive: line 25
Never mutate live state directly.
- Very specific: line 63
After approval, stage autoevo_pending.py resolve --id <id> --status applied --reason <reason> --today <date> on private state.
- Very specific: line 68
If the user rejects the proposal, stage a dismissed resolution with their reason instead.
- Distinctive: line 95
Stop. Write only already confirmed staged changes.
- Very specific: line 100
Before writing, recheck live hashes and limit the write to approved note/state/review-log paths.
- Distinctive: line 117
Do not replay the model, automatically roll back, or overwrite state.
≈ “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
/autoevo-review
Triage pending Autoevo proposals with the user: apply, skip, defer, or explain.
Follow protocols/autoevo.md. This is an interactive approval workflow, not a
continuation of the nightly model run.
Never edit canonical nightly JSON or its derived reports during review; doing so
cannot prove an incomplete cycle. Optional human history belongs in a separate
<paths.agent_findings>/autoevo-review-<YYYY-MM-DD>.md.
1. Load and housekeep
Resolve <paths.meta>/autoevo_pending.toml and the current local date once.
Load through scripts/autoevo_pending.py list --status pending; a missing queue
or zero returned entries means there is nothing to review. A malformed queue is
a blocker: report it and do not rewrite it.
Before presentation, stage autoevo_pending.py auto-dismiss --today <date> on
private state for entries at three surfaces or over the default 30-day age.
Never mutate live state directly. After approval, the orchestrator writes explicit note, queue, ledger, and optional review-log changes as plain files after rechecking their hashes; Reflect commits and pushes them.
2. Present the queue
Group remaining entries by redundant, time-stale-A, time-stale-B,
contradicted, and low-signal. Show a compact category count followed by each
entry's id, action, evidence, peers, proposal date, surface count, and any
default_action / default_at.
Explain veto semantics exactly:
- Skip vetoes a
stale-bannerdefault. - Apply vetoes a
dismissdefault; skip confirms it. - Defer restarts the 14-day window.
- Silence leaves an eligible default to the later trusted nightly parent.
Let the user choose a category or all entries. If unclear, walk all entries in order. For each item ask: apply, skip, defer, explain, or quit. Apply and skip require one user-authored sentence of reason; never invent or accept an empty reason. Defer may carry an optional reason.
3. Resolve one item
Apply
Dispatch the relevant normal, user-approved workflow. Curator drafts note
changes and retains its full preservation/approval gates. Autoevo's
auto_apply_safe flag does not bypass approval here.
A queued contradiction is not necessarily confirmed: the nightly queues both genuine and unproven contradictions. Unless the entry carries a complete, gap-free Challenger verdict tied to the exact claim and peer, dispatch Challenger first. A genuine verdict may proceed to Curator's normal wiki-update proposal; a rhetorical verdict performs no wiki write and returns to the user for a dismissal decision.
After approval, stage autoevo_pending.py resolve --id <id> --status applied
--reason <reason> --today <date> on private state. Write it with the approved
note change. Archive means byte-preserving addition plus deletion, not a model
write or promised rename.
If the user rejects the proposal, stage a dismissed resolution with their
reason instead. No content change is written.
Skip
Stage autoevo_pending.py resolve --id <id> --status dismissed --reason
<reason> --today <date> against private state. Tell the user whether this
vetoes stale-banner or confirms dismiss; then include it in the next queue
write.
Defer
Stage autoevo_pending.py defer --id <id> --today <date> with the optional
reason. It increments surfacing state and, when applicable, restarts the default
deadline. Optionally ask for a cue snooze (default seven days); cue snoozing is
group-wide, not per entry. Batch compatible defer-only state changes into one
write.
Explain
Show the complete stored evidence without adding authority. For redundant QMD findings, show peer paths, QMD ordering scores, and source-reviewed overlap; do not describe a score floor. For contradictions, show the claim, peer, signal, probe verdict, and any unresolved gaps. Then ask again.
Quit
Stop. Write only already confirmed staged changes; leave every other entry pending. If writing has not begun, discarding scratch state has no live effect.
4. Write and summary
Before writing, recheck live hashes and limit the write to approved
note/state/review-log paths. Never commit or push. Rollback is plain
git restore.
Report applied, dismissed, deferred, remaining, and any failed write. Name the queue path and separate review log if one was created. A review log is ordinary human-triage history only; it is never an Autoevo verifier input.
Edge cases
- Missing peers or changed evidence make an entry stale. Explain the mismatch and ask the user whether to dismiss; do not silently resolve it.
- A wiki change always uses Curator's normal approval path and appends the required Revision Log.
- Any daily-note source is a nightly bug. Refuse the content operation and ask the user whether to dismiss the queue entry; daily notes remain read-only.
- On ambiguous or partly written effects, stop and report
needs_review. Do not replay the model, automatically roll back, or overwrite state.
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.

