From forge
Run verification commands and show output before any completion claim. Evidence before assertions. Fresh execution required, not cached results.
npx claudepluginhub caseyrtalbot/forge --plugin forgeThis skill uses the workspace's default tool permissions.
Verify that the entire body of work is complete, correct, and ready to integrate. This is the final checkpoint before code lands. Every claim of completion must be backed by executed evidence, not assertions.
Verifies tests pass on completed feature branch, presents options to merge locally, create GitHub PR, keep as-is or discard; executes choice and cleans up worktree.
Guides root cause investigation for bugs, test failures, unexpected behavior, performance issues, and build failures before proposing fixes.
Writes implementation plans from specs for multi-step tasks, mapping files and breaking into TDD bite-sized steps before coding.
Verify that the entire body of work is complete, correct, and ready to integrate. This is the final checkpoint before code lands. Every claim of completion must be backed by executed evidence, not assertions.
"I wrote it correctly" is NOT evidence. "The tests should pass" is NOT evidence. "I checked and it looks fine" is NOT evidence.Evidence is OUTPUT. Run the command. Capture the result. Show the result. Only then can you claim completion.
Verification must be a FRESH execution. Cached results, prior session output, or "I already ran this" are not evidence. Run it now, show the output now.
Do NOT output any completion claim, success message, or "all done" statement until you have run verification commands and can show their output.
digraph confirm_complete {
"Read workflow state" [shape=box];
"Run full test suite" [shape=box];
"Tests pass?" [shape=diamond];
"Run build" [shape=box];
"Build passes?" [shape=diamond];
"Run security scan" [shape=box];
"Security clean?" [shape=diamond];
"Spec coverage audit" [shape=box];
"All specs covered?" [shape=diamond];
"Record evidence" [shape=box];
"Invoke land-changes" [shape=doublecircle];
"Return to drive-execution" [shape=box];
"Read workflow state" -> "Run full test suite";
"Run full test suite" -> "Tests pass?";
"Tests pass?" -> "Run build" [label="yes"];
"Tests pass?" -> "Return to drive-execution" [label="no"];
"Run build" -> "Build passes?";
"Build passes?" -> "Run security scan" [label="yes"];
"Build passes?" -> "Return to drive-execution" [label="no"];
"Run security scan" -> "Security clean?";
"Security clean?" -> "Spec coverage audit" [label="yes"];
"Security clean?" -> "Return to drive-execution" [label="no"];
"Spec coverage audit" -> "All specs covered?";
"All specs covered?" -> "Record evidence" [label="yes"];
"All specs covered?" -> "Return to drive-execution" [label="no"];
"Record evidence" -> "Invoke land-changes";
}
Deploy the integration-verifier agent to run the project's complete test suite (not just the tests for changed files). Capture the output. All tests must pass.
Deploy the test-strategist agent to audit test coverage. Verify that all spec requirements have corresponding tests and that no critical paths are untested.
Run the project's build command. Capture the output. Build must succeed with no errors. Warnings should be noted but do not block.
Deploy the security-sentinel agent to scan the diff for:
Read the original spec document. For each requirement, verify there is:
Write verification results to .forge/evidence/verification/:
test-results.txt -- full test suite outputbuild-results.txt -- build command outputsecurity-scan.md -- security sentinel findingsspec-coverage.md -- requirement-to-task mapping with verification status"All tests pass" (without showing output) Show the output. "All tests pass" is a claim. The terminal output showing "47 tests passed, 0 failed" is evidence.
"I ran the build in my head" Run it for real. Mental execution misses environment issues, missing dependencies, and configuration problems.
"Security is probably fine, it's an internal tool" Internal tools get promoted to external tools. Vulnerabilities in internal tools become vulnerabilities in external tools. Scan it.
| Claim | What's actually required |
|---|---|
| "Tests pass" | Show terminal output: "N tests passed, 0 failed" |
| "Build succeeds" | Show build output with exit code 0 |
| "No security issues" | Show security scan report |
| "All requirements met" | Show spec-to-task mapping with each verified |
| "Agent completed the work" | Show agent's output, verify claims independently |
If all checks pass: invoke land-changes to integrate the code. If any check fails: return to drive-execution to remediate.