Verification Before Completion
Makes the agent check evidence before claiming tests pass, a build works, or a bug is fixed.
- Repository
- mvschwarz/openrig
- Stars
- 8,553
- Licence
- Apache-2.0
- Files
- 3 files
What it does
This is a set of working rules that makes an AI agent back up its 'it's done' or 'it works' statements with actual evidence, such as a test run, a build exit code, or a checked artifact, before it says so. It tells the agent to state whether it ran a check itself or relied on someone else's result, to match the depth of checking to the stakes, and to re-run checks when inputs change. It also lists common failure patterns, such as treating a linter pass as proof the code compiles, and warns about checks that look like success while proving nothing.
Great for
- Developers who want an agent to verify its own code changes before committing or opening a pull request
- Teams reviewing agent-written work who want clearer reporting of what was actually checked
- People building or customizing their own agent skills who want a model for evidence-based completion rules
Skip it if
- Quick exploratory or throwaway tasks where full verification would slow things down without benefit; the skill itself says to scale checks to stakes
- Non-verifiable work such as creative writing or opinions, where there is no test or build result to check
- Anyone expecting it to run tests or tools by itself: it is guidance text that shapes the agent's behavior, and whatever checks run depend on the project and the agent's environment
Try asking
- “Before I open this pull request, make sure the tests actually pass and tell me what you ran.”
- “Fix this bug and confirm the original failing test now passes before saying it's fixed.”
- “The subagent says the migration succeeded; check the output file before we report that as done.”
- “Write a regression test for this bug and show me it fails without the fix.”
Install in 3 steps
- 1
Download the skill
The ZIP contains the 3 original files, the Apache-2.0 licence and an attribution note.
verification-before-completion.zip - 2
Add it to your agent
Claude Code (all projects)
mkdir -p ~/.claude/skills && unzip verification-before-completion.zip -d ~/.claude/skills/Claude Code / Agent SDK (this project only)
mkdir -p .claude/skills && unzip verification-before-completion.zip -d .claude/skills/Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.
- 3
Ask for the task
No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.
Get better results
- Name the specific command or artifact you want checked (for example, 'run npm test') so the agent has a concrete check to run and report.
- Ask the agent to say whether it ran a check or relied on another agent's report, since the skill asks for that distinction.
- For bug fixes, request a red-green cycle (test fails without the fix, passes with it) to confirm the test actually catches the bug.
What's inside
- LICENSE.txt
- references/performance-claims.md
- SKILL.md
Redistributed unmodified from mvschwarz/openrig under the Apache-2.0 licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.