agentsclimarketplace

Finish the loop

Skill zakelfassi/skills-driven-development/examples/proof-carrying-skill/finish-the-loop

Agents that learn by doing — and remember how they did it. A methodology for AI agents to create, evolve, and share reusable skills.

Install
npx -y skills add zakelfassi/skills-driven-development --skill finish-the-loop

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 17 stars17 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Close implementation loops with target-bound independent evidence and a portable proof receipt. Use when an agent changes a real system and must earn a completion claim that survives the session.

SKILL.md

1.4 KB, as published. Nobody here has run it

Finish the Loop

Treat completion as a claim that needs evidence from the intended target.

  1. Read WORKFLOW.json and create .skdd/runs/<run-id>/state.json before changing the system.
  2. Pin each acceptance line to an explicit target tuple: environment, resource, and revision. Do not let the executor silently choose or revise the target.
  3. Record the executor identity and credential scope.
  4. Reproduce, change, and drive the real artifact. Keep the run state resumable after interruption.
  5. Collect evidence with a provider that has a different identity and credential scope from the executor. A test run by the same process is useful feedback, but it is not independent proof.
  6. Emit receipt.json using the proof-receipt schema. Bind the skill bytes and each acceptance contract with SHA-256 digests.
  7. Run skdd proof verify receipt.json --skill SKILL.md.
  8. Report verified only when the verifier passes. Otherwise report the reason codes and leave the loop refused, blocked, or open.

Never translate “the build passed” into “the user-visible outcome works.” Evidence must match the acceptance target exactly.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.