configs/claude-code/skills/verify-claims/SKILL.md
Use when about to claim work is complete or passing, when receiving subagent results, or when writing a regression test for a bug fix
npx skillsauth add jimweller/dotfiles verify-claimsInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
STARTER_CHARACTER = 🧾
Three procedures for proving claims with evidence. Each procedure runs before the claim is made, not after.
Map the claim to the command that proves it. Run the command in this turn. Use the actual output, not memory of prior runs.
| Claim | Command | Insufficient | | --------------------- | ---------------------------------------------- | ------------------------------ | | Tests pass | Project test command, full suite or named test | Prior run, "should pass" | | Linter clean | Project linter command | Partial check, extrapolation | | Build succeeds | Project build command, exit 0 | Linter passing, logs look fine | | Bug fixed | Test that reproduces the original symptom | Code changed, "looks fixed" | | Regression test works | Red-green-revert-restore (see below) | Test passes once | | Subagent completed | VCS diff plus targeted re-check | Subagent self-report | | Requirements met | Re-read spec, line-by-line checklist | Tests passing |
Subagents report success in their final message. The report is not evidence. The diff is.
When a subagent returns:
git status and git diff to see what actually changedSubagents fabricate success reports under pressure. Treat the report as a hypothesis to verify, not a fact to accept.
A test written after the fix passes immediately. Passing immediately proves nothing. To prove the test catches the bug:
If step 4 passes, the test does not exercise the bug. Throw it out and write a new one.
This procedure is unnecessary when test-driven-development order is followed (test first, then fix). It is the recovery procedure for the inverted order.
testing
Search saved session transcripts for past decisions, actions, errors, and context that has left the current conversation window.
data-ai
Review a PRD for defects via Claude opus subagent.
development
Markdown authoring guidelines for formatting, code blocks, and structure. Use when writing or editing markdown files.
development
--- name: md-style description: README style guide for concise, direct documentation. Use when writing or editing README files. ß--- <!-- markdownlint-disable-file MD041 --> # README Style Guide Write concise, direct README files for experienced engineers. ## Principles - **No fluff** - Skip tables of contents, verbose explanations, development history - **No roadmaps** - Document current state only, not plans or decisions. Readme is an engineering specification. Not a project plan or chan