Verify, analyze, converge

Proof that work is done lives in git: a content hash in frontmatter, optional JUnit, and covering tasks for each criterion.

Verify

pb verify TASK-001
pb verify TASK-001 --force

1. Runs each string in checks.commands as an argv (no shell interpolation).
2. Optionally reads checks.report (repo-relative JUnit XML) afterward.
3. Stamps verified: { at, checks, hash } on the item.
4. Reloads and refreshes the board.

--force sets bypassed: true when you must skip failing checks. Write down why.

--json includes results ({ name, classname, status, time } per test) and reportStale when the JUnit file was not rewritten by this run. A missing or corrupt report is not an error — results is empty.

If checks.commands is empty, verify still stamps a hash.

Trust boundary

Commands run at the same trust level as package.json scripts. They are not sandboxed. Do not point checks.commands at untrusted input. See SECURITY.md.

Config

checks:
  commands: [pnpm test, pnpm lint]
  report: .pb/junit.xml

When commands are configured, status: done without a verified block is lint error unverified-done. Editing the item after verify is stale-verified until you verify again.

Analyze

pb analyze
pb analyze --json

Coverage table without an LLM:

Column	Meaning
Requirement Key	Story/epic criterion or a rule/ADR id
Has Task?	A non-cancelled task lists that key
Task IDs	Covering tasks
Proved?	For ID#N rows: a passing JUnit test mentions that exact token
Test	Matching test name
Notes	Gaps

• coveragePercent — share of rows that have a covering task
• provedPercent — share of acceptance-criteria rows with a passing bound ID#N test
• JSON adds proved / unproven arrays of { id, index, test?, status? }

Matching is the exact token ID#N in JUnit classname + " " + name. fail, error, and skipped count as unproven. Rule and ADR rows are not machine-ownable.

Exit 1 when the graph is not ok (uncovered active rules, or done stories with open children).

Name tests after the criterion they prove:

it("US-001#2 rejects a float amount", () => { /* … */ });

Converge

pb converge US-001 --dry-run
pb converge US-001
pb converge EPIC-001

Appends tasks for acceptance criteria that have no covering covers: [ID#N] task. --dry-run prints a plan (converged or plan + titles). It does not rewrite existing tasks.

Covers

On a task:

covers: [US-001#1, US-001#2]

N is the 1-based index of checkboxes under ## Acceptance criteria on that story (or the target id). See ADR-0007.

Bump

pb bump BR-001
pb bump ADR-0001

Increments version, sets amended to today, refreshes content_hash. Use after you change the body of an active rule or accepted ADR. If the body is unchanged, you get a warning instead of a bump.

Impact

pb impact BR-001

Lists stories and tasks that cite that rule or ADR (inbound edges), with status and whether they are done.