agentleFS
youcislayoucisla/Agent-Foundry/skills/core/quality-protocol/SKILL.md

quality-protocol

Skill1 starsChanged 23 days agoMITView on GitHub

The routing protocol for non-trivial work: restate, catalog constraints, choose one discipline, execute in verified increments. Use before declaring a non-trivial task complete.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 3 are distinctive, and 0 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.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Add the success criteria: how a reviewer can tell the task is done.

    line 34
  2. Distinctive:

    Use one entry point. Stacking every discipline turns the protocol into ceremony and buries the actual constraint.

    line 47
  3. Distinctive:

    Check every constraint from step 2.

    line 63

≈ “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

Quality Protocol

An umbrella for non-trivial work. It does not repeat every planning and verification procedure; it chooses the one that fits and keeps the work honest.

Procedure

1 - Restate the request

One short paragraph: goal, known context, gaps, assumptions. Skip the ceremony when the request is already unambiguous; still name the assumption you act on.

2 - Catalog binding constraints

List only constraints that can fail the work:

  • Technical: platform, version, compatibility
  • Scope: files or surfaces allowed to change
  • Evidence: what must be checked against a live source
  • Boundaries: read-only, no commit, no credentials, no destructive action

Add the success criteria: how a reviewer can tell the task is done.

3 - Choose one entry discipline

SituationDiscipline
Ambiguous request, unclear constraintsconstraint-then-solve
Non-trivial change with more than one reasonable approachplan-before-code
A plan exists and execution order mattersplan-then-act
A claim must be checked before it is repeatedverify-first
An inherited finding or audit item is about to be acted onre-verify-findings
A design document is being implementedread-before-build

Use one entry point. Stacking every discipline turns the protocol into ceremony and buries the actual constraint.

4 - Plan and take the first action

State the plan in one sentence and the first concrete step. If the user already authorized the work and the spec is clear, proceed; do not re-ask for approval.

5 - Execute in small verified increments

For each increment: change, run the narrowest check, inspect the result, then continue. A change is not done because it was made; it is done when its check passes. Batch independent inspections, serialize dependent changes.

6 - Verify against the catalog

Check every constraint from step 2. Mark each one pass or fail with the evidence used. If a constraint cannot be checked, say so; do not imply it passed.

7 - Close with evidence

Report what changed, what was checked, and what remains uncertain. Do not claim completion from the change alone.

Skill boundaries

  • plan-before-code owns the spec and approval gate.
  • plan-then-act owns execution cadence.
  • constraint-then-solve owns ambiguity and constraint reasoning.
  • verify-first owns checking a new assertion.
  • re-verify-findings owns inherited findings.
  • read-before-build owns source inspection before implementation.

Anti-patterns

  • Running every discipline on a one-line change.
  • Restating the request and then ignoring the constraints during execution.
  • Treating a passing compile as proof that the requested behavior works.
  • Claiming a constraint passed without naming the evidence.
  • Asking for a second approval the user already gave.

References

  • details

Verification Checklist

  • The success criteria are stated in observable terms.
  • One entry discipline was chosen and named.
  • Each binding constraint is marked pass, fail, or unverified.
  • The final report separates what changed from what was checked.
  • Remaining uncertainty is stated rather than hidden.

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.