agentleFS
omidnwomidnw/my-skills/skills/reliable-execution/SKILL.md

reliable-execution

Skill0 starsChanged 10 days agoMITView on GitHub

Execute well-scoped non-trivial tasks the way the user expects — plan before acting, stay in scope, verify before reporting. Load once the task is sufficiently specified; if the request is vague, under-specified, or missing required context, invoke task-intake first. Also use when a session is long or a model tends to drift. Forces task restatement, step decomposition, scope checks, and honest reporting so the delivery matches the request.

The file

Show its rules and text

We split this skill into the rules it gives an agent

14 rules. About 0 are common, 12 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈12

    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
  1. Distinctive:

    Before starting any task that involves more than a one-line change.

    line 25
  2. Very specific:

    When a well-specified request is large or spans multiple files (for vague or under-specified requests, run task-intake first).

    line 26
  3. Distinctive:

    When the session is long and drift risk is high.

    line 28
  4. Very specific:

    Restate the task in one sentence: If you cannot restate it, re-read the request before doing anything.

    line 35
  5. Distinctive:

    Name the deliverable: Write down what "done" looks like.

    line 36
  6. Distinctive:

    Execute in scope: Only what was asked. Note unrelated findings separately.

    line 39
  7. Distinctive:

    Re-read the original request: Check off every requirement.

    line 40
  8. Distinctive:

    Verify: Run the relevant checks (typecheck, lint, tests, or a manual test of the affected flow).

    line 41
  9. Distinctive:

    Never add unrequested work to the delivery.

    line 47
  10. Distinctive:

    Prefer the smallest change that satisfies the requirement.

    line 48
  11. Distinctive:

    Keep restatement, acceptance criteria and planning internal unless a plan is requested.

    line 49
  12. Distinctive:

    Never present unverified work as done.

    line 51
  13. Distinctive:

    Every "done" claim is backed by a verification step.

    line 63
  14. Distinctive:

    No file outside the request was modified.

    line 64

≈ “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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Reliable Execution — Universal

Persistent design workflow companion

For complex feature/screen/API/architecture design, consider design-to-delivery as an adaptable playbook. Its diagram → brainstorming → docs/specs/<work-id>/specs.md → docs/task/<work-id>/tasks.md sequence helps expose missing states, contract inconsistencies and verification gaps. Choose or simplify the useful parts; ordinary design requests do not require every artifact. Explicitly requested documents remain deliverables and the internal-planning rule below does not suppress them. Continue authorized implementation using the selected plan without duplication. Preserve all existing documents and history.

Purpose

Turn any request into an on-scope, verifiable delivery. Especially valuable for models that tend to drift mid-task: it forces restatement, decomposition, scope discipline, and honest reporting.

When to use

  • Before starting any task that involves more than a one-line change.
  • When a well-specified request is large or spans multiple files (for vague or under-specified requests, run task-intake first).
  • When the session is long and drift risk is high.

Preconditions

  • The original request is available to re-read (do not rely on memory of it).
  • Verification commands for the project are known or discoverable (package.json scripts, Makefile, CI config).

Workflow

  1. Restate the task in one sentence. If you cannot restate it, re-read the request before doing anything. For vague or under-specified requests, run the task-intake skill first to build the brief.
  2. Name the deliverable. Write down what "done" looks like.
  3. Decompose into steps. Break the work into the smallest number of concrete steps that cover the whole request; order them.
  4. Gate each step. Execute work already authorized by the request, including large reversible tasks. Ask only for destructive/irreversible actions, missing required authorization, blocking ambiguity, or a materially broader outcome; size alone is not an approval gate.
  5. Execute in scope. Only what was asked. Note unrelated findings separately; do not fix them.
  6. Re-read the original request. Check off every requirement; complete anything missing.
  7. Verify. Run the relevant checks (typecheck, lint, tests, or a manual test of the affected flow).
  8. Report. What changed, what was verified, what failed or is uncertain — in that shape.

Rules

  • Work like a maintainer: derive code style from explicit project conventions and maintained approved examples, including naming, module boundaries, errors and test/helper style. Preserve understandable diffs and existing behavior; do not imitate known defects or claim to know unrecorded personal preferences. Use proportional self-review before handing meaningful changes back. Flexible process is not a reason to omit care or hide verification gaps.
  • Ambiguity that changes the outcome → ask. An obvious, safe, reversible default → proceed and state it.
  • Never add unrequested work to the delivery; report it as a separate note.
  • Prefer the smallest change that satisfies the requirement; preserve existing behavior.
  • Keep restatement, acceptance criteria and planning internal unless a plan is requested. Preserve useful existing content when improving instruction files; do not replace an improvement request with a shortening campaign.
  • Before edits, read the current target and related contracts; afterward re-read and inspect the diff. For behavior changes use behavioral-testing, for E2E use the project skill or e2e-test-authoring, for non-trivial self-review use engineering-review, and for static checks use code-quality. These are companions, not duplicate execution plans.
  • Never present unverified work as done. If you could not verify, say so plainly.
  • Reversible bias: prefer additive changes over replacements; backup before irreversible operations.

Common mistakes

  • Starting to code before restating the task — this is where drift begins.
  • "Also fixing" unrelated issues noticed along the way.
  • Reporting success without running verification.
  • Asking the user when the default is obvious — wastes their time.
  • Expanding scope to make the solution "nicer" instead of delivering what was asked.

Validation

  • The final report maps 1:1 to the original request.
  • Every "done" claim is backed by a verification step.
  • No file outside the request was modified.

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.