agentleFS
youkistudiosyoukistudios/honest-loop/skills/honest-cycle/SKILL.md

honest-cycle

Skill2 starsChanged 4 months agoMITView on GitHub

The governed loop — /honest, /push, /honest again, until 9/10 or a principled stop. Fires on "honest cycle", "run the loop", "quality loop", or after finishing any phase of planned work. Use --plans for a batch of plans (sequential, each reaches 9 before the next starts). Ends with an independent fresh-context verify — a model doesn't grade its own homework.

  • In this repository2 other files it gives its agents.

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:

    Before anything is declared done at ≥9: a fresh-context agent (no memory of the build or the pushes) re-runs /honest on the subject and …

    line 39
  2. Very specific:

    No subagents available → the self-score stands but is labeled provisional — self-assessed wherever it's recorded, and the first fresh …

    line 42
  3. Distinctive:

    every phase has a verifiable "done when"

    line 60

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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

/honest-cycle — the loop with brakes

One command: score → push → re-score, repeating toward 9, with hard exits so it can never become polishing-forever or grade-inflation. Vocabulary lives in /honest.

The loop

[1] /honest ──► ≥9 ──────────────► [4] independent verify ──► done
      │                                     ▲
     7–8 ──► [2] /push ──► [3] /honest ──► ≥9
      │                        │
      │                       7–8 ──► one more push (the second and LAST)
      │                        │
     ≤6                       ≤6 or score dropped
      │                        │
      ▼                        ▼
  STRUCTURAL STOP          STRUCTURAL STOP / REGRESSION STOP

Exit conditions (the brakes)

ConditionMeaningAction
Score ≥ 9 twice (self + independent)genuinely donelog, ship
≤ 6 at any pointstructural — pushing polishes a wrong thingSTOP; attach the structural diagnosis (wrong-problem/-shape/-owner/-time) and its prescription; wait for the human
Below 9 after 2 pushesgaps are symptomsSTOP; name the structure, never push a third time
Score dropped after a pushthe push made it worseSTOP; review the last commit before anything else
VALIDATE gaps remainneeds a real run only the user can dosurface exact test; continue other work
DECISION gaps remainneeds a human callsurface exact question; continue other work

[4] Independent verify — the default, not the exception

Before anything is declared done at ≥9: a fresh-context agent (no memory of the build or the pushes) re-runs /honest on the subject and reports its own score.

  • Harness has subagents → dispatch one; that score is the final one.
  • No subagents available → the self-score stands but is labeled provisional — self-assessed wherever it's recorded, and the first fresh session to touch the subject should confirm it.
  • Independent score disagrees by ≥1 point, or finds a gap self-review missed → that disagreement is itself a gap; loop once more (doesn't count against the 2-push cap — it's new information).

Why this is the default: a real self-scored 9 was once independently re-read and found to be a 7 with fourteen broken files. The self-assessment wasn't dishonest — it checked everything it thought to check. The miss was structural: nobody outside the room had looked. Now someone always does.

--plans (batch mode)

For N plans: strictly sequential — plan 1 reaches 9 before plan 2 is scored. Batch pressure ("finish the set") is the enemy: parallel assessment produces shallow gap lists. Additional plan-specific checks during [1]:

  • success criteria test behavior, not document existence
  • every phase has a verifiable "done when"
  • technical choices name at least one rejected alternative
  • automation steps don't assume they can do human-participation work
  • routing/decision rules exist (no "human decides" hand-waving) A plan that can't reach 8 after one push has a structural problem — stop the batch, surface it; never approve a plan to keep a batch moving.

Output per cycle

✅ /honest-cycle — [subject]
[pre]/10 → [post]/10 (independent: [score]/10 | provisional)
Gaps closed: n · Deferred: [VALIDATE/DECISION items with exact commands/questions]
Status: DONE | STRUCTURAL STOP ([diagnosis]) | REGRESSION STOP | HUMAN NEEDED

Optional trajectory line as defined in /honest — and when the independent verify lands, write the confirmed score back: append a final line with "provisional": false superseding the self-assessed entry (append, never edit — the log stays append-only). Delta checks read the latest line per subject.

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.