mitodl/agent-kit/verify-before-claiming
Claude Code and agent skill by mitodl/agent-kit
- community
Checking that the SKILL.md file is still in the repository.
Use this skill before stating the status of anything outside the conversation: a PR, CI run, review, witan task, branch, deploy, environment, or fix. Triggers on "what's the status of", "is it merged", "did CI pass", "did the review land", "is it deployed", "is it fixed", "is X done", "where are we on", and on any summary or handoff that reports status. Re-fetch live state instead of repeating memory or earlier output, check uncommitted work and other worktrees before calling work missing, and…
Install
npx vskill@latest install mitodl/agent-kit/verify-before-claimingSKILL.md
Verify before claiming
State changes while you work. What you saw earlier in this conversation, or what a summary or memory says, is not current state. These rules apply to every status claim, not only to code review.
- Re-fetch status in the same turn you state it, and declare an environment or
fix done only after running the check that shows it works.
- PR:
gh pr view <n> --json state,reviewDecision - CI:
gh pr checks <n> - witan task:
task_get(slug=...) - Branch:
git fetch && git log origin/<branch> -1
- Never report an external event (a review landed, a check finished, someone
replied, a deploy completed) without the API call that shows it.
- Before concluding work is missing or not started, run
git statusand
git worktree list. Uncommitted changes and other worktrees count.
- Label each finding you report "verified (
<command>)" or "inferred".
Anything you did not check is inferred, and you say so.
Shell commands and file edits
The shell may be bash, zsh or fish, and the tools GNU or BSD.
- Anything beyond a single pipeline or
&&chain goes in a script file
outside the repo (the session scratchpad if your harness gives one, else mktemp -d), run with bash <file> or python3 <file>. No inline loops or multi-line heredoc logic in the interactive shell.
- Don't depend on one shell's behavior in a one-liner. Quote every variable.
Don't assign to names zsh reserves (status, path). GNU and BSD tools differ (sed -i, date, grep -P), so do in-place edits of generated or data files in python, not sed -i.
- Edit source files with your edit tool (Edit,
apply_patch), never with
sed -i, a heredoc, or a python rewrite. After any scripted change to a file, read the result back.
- Strip CR (
tr -d '\r') from a file that may have CRLF line endings before
parsing it with awk, cut, read or a $-anchored grep.
ps | grepmatches itself, andpgrep -fmatches the shell wrapping your
command. In wait loops, save $! and poll kill -0 "$pid", or use a pidfile.
- Commands handed to the user to paste use real values, never placeholders,
and say which shell they are for. Secrets go through an env var ("$GITHUB_TOKEN"), never pasted into the command.
Security
- Security scanPassed, 100 of 100Pattern checks on the skill's files.
- Model reviewNot runA model review is scheduled for the next rescan.
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/workflow/verify-before-claiming/SKILL.md
Popularity
- Repository stars
- 5
- 7-day trend score
- 0
Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/mitodl/agent-kit/verify-before-claiming)