agentsclimarketplace

Verify before done

Skill stevennitesh/programming-agent-skills/skills/.archive/verify-before-done

Use before claiming repo work is done, fixed, passing, reviewed, ready, resolved, clean, safe to merge, mergeable, or complete; calibrates claim strength to fresh evidence.From its SKILL.md

Install
npx -y skills add stevennitesh/programming-agent-skills --skill verify-before-done

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

  • 0 stars0 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.

SKILL.md

3.8 KB, 818 tokens by cl100k_base, as published. Nobody here has run it

Verify Before Done

Calibrate completion and readiness claims to fresh repo evidence. This is a gate, not a semantic review workflow.

Rule

Before saying done, fixed, passing, reviewed, ready, resolved, clean, safe, mergeable, or complete, write the exact claim and prove only that claim.

First visible move:

Claim:
Final edit point:
Evidence after final edit:
Diff checked: yes|no
Supported wording:

Narrow evidence supports narrow claims. If evidence is weak, stale, partial, simulated, missing, or does not cover the touched behavior or risk, downgrade the wording.

Do not turn "I ran a command" into "the work is done." The check, diff review, CI/PR state, review state, or manual observation must support the exact claim.

Do not use this skill to perform semantic PR review or review-comment resolution. Route that work through coding-router.

Fast Claim Check

Use for tiny edits, narrow reports, doc-only changes, read-only answers, and small local fixes.

  1. Write the exact claim.
  2. Name the final edit point, or none for read-only work.
  3. Name evidence gathered after the final edit: command, test, diff inspection, source read, CI/PR state, review state, or manual step.
  4. Check the diff when local files changed.
  5. State the strongest supported wording.

For small work, one line is enough:

Verified: <claim> supported by <evidence>; remaining risk: <risk|none named>.

Full Gate

Use for behavior changes, PR/merge readiness, generated/config/workflow changes, broad cleanup, public or caller risk, or when the requested claim is stronger than local evidence.

  • Inspect the final diff against the request and scope; separate unrelated dirty work.
  • Run the smallest sufficient check after the final edit; say what it covers and what it does not cover.
  • Check CI, PR, branch, mergeability, and review-thread state only when that state is part of the claim.
  • Name skipped checks, warnings, flakes, unrelated failures, manual-only evidence, and residual risk.
  • Downgrade unsupported claims.

Diff Review

When files changed, inspect the diff for:

  • changes outside the request or accepted scope
  • behavior outside scope, missing acceptance coverage, or weak test surface
  • evidence that does not exercise the touched path or stated risk
  • unrelated user work, generated output, config, dependency, migration, or workflow changes
  • failed, skipped, stale, simulated, or pre-final-edit evidence

Diff review is not a substitute for pre-pr-review. If semantic review is the job, return to coding-router.

Stop Or Downgrade

Stop, reroute, or downgrade the claim when:

  • the final edit happened after the last relevant check, diff review, CI/PR state check, or manual observation
  • the check does not cover the touched behavior, public or caller contract, source path, config/workflow change, generated output, or named risk
  • the diff was not inspected
  • CI, review, branch, merge, or PR state matters but was not checked
  • evidence is simulated, stale, manual confidence, or subagent-only
  • failures, warnings, flakes, or unrelated changes are unexplained
  • the requested wording is stronger than the evidence supports

Handoff

If proving the claim reveals new work, return to coding-router.

Use pre-pr-review for semantic diff review, diagnose-loop for unexplained failures, workspace-safety for overlapping dirty paths or Git risk, github-tracking for durable issue/PR records, and GitHub plugin skills or tools for live PR/comment/CI state.

Report

For small work, use the one-line Fast Claim Check report.

For nontrivial work:

Claim:
Evidence:
Diff review:
Unsupported or downgraded claims:
Remaining risk:

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Keep looking

Skills are one crate of 326,835. 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.