Verification Before Completion

Core principle: Evidence before claims, always.

When To Apply

MUST apply before:

• Any commit or push operation
• Any completion claim ("done", "finished", "complete", "fixed")
• Any summary of completed work to the user
• Moving to the next task after finishing one

Verification Protocol

Before claiming any status:

1. Detect: What build/test command does this project use?
2. Run: Execute the command (fresh, complete)
3. Read: Full output, check exit code, count failures
4. Confirm: Does output confirm success?
   • If yes: State result with evidence. Proceed.
   • If no: Read the diagnostic output. Fix the problem. Re-run from step 2.
   • If tool unavailable: Stop and tell the user.
5. Only then: Make the claim, commit, or declare done

Skip any step = no claim permitted.

When Verification Cannot Run

If the verification command is not available in the environment
(e.g., nix: command not found),
this is not an exemption.
Do not label the work "UNVERIFIED" and proceed anyway.
Stop and tell the user the verification command is unavailable.

Detecting the Verification Command

Check the project root for these files and run the corresponding command:

File	Command
flake.nix	nix flake check or nix build
go.mod	go build ./... and go test ./...
Cargo.toml	cargo build and cargo test
build.zig	zig build test
package.json	npm test
Makefile	make check or make test
pyproject.toml	pytest

If multiple apply, run all that are relevant to the changed files.
If none apply, state explicitly: "No build/test command found for this project."
Do not treat absence of a build tool as permission to skip verification.

Red Flags

Stop immediately if you are about to:

• Say "should", "probably", "seems to" instead of running verification
• Express satisfaction before verification ("Great!", "Perfect!", "Done!")
• Commit or push without running the build/test command
• Claim "my tests pass" when the full suite has failures
• Say "looks correct" based on reading code instead of executing it
• Rationalize skipping verification ("simple change", "config only", "just docs")

Common Failures

Claim	Requires	Not Sufficient
Tests pass	Full suite: 0 failures	"My new tests pass"
Build succeeds	Build command: exit 0	Reading the diff
Bug fixed	Test reproducing symptom: passes	Code changed
Config works	Config evaluates without error	"Syntax looks valid"

Verification Patterns

Build:

Good: Run `nix flake check` -> exit 0 -> "Build passes"
Bad:  Read the diff -> "Looks correct" -> commit

Tests:

Good: Run full test suite -> 34/34 pass -> "All tests pass"
Bad:  "My new tests pass" (ignoring pre-existing failures)

Bottom Line

Run the command. Read the output. Then claim the result.

No exceptions.