Execution Log: six graph nodes gate on execution-log sections without declaring an artifact

Append-only log of progress for the user's visibility. Checked in alongside the spec at docs/specs/issue-167/execution-log.md.

Phase transitions

Phase	Entered	Reviewed/approved by	Notes
requirements-definition	2026-08-06	pending (PR)	filed as a bugfix; written as requirements.md because the fix adds vocabulary, not only a repair
design	2026-08-06	pending (PR)	option 2 (validates:) + option 3 (fail closed); option 1 rejected — decision-063
test-planning	2026-08-06	pending (PR)	reviewed with the design, one gate for both (decision-060 D2)
tasks-breakdown	2026-08-06	pending (PR)	7 tasks: T1/T2 → T3 → T4/T5 → T6 → T7
implementation	2026-08-06	—	T1–T6
verification	2026-08-06	—	T7; every activity ran
needs-review	2026-08-06	pending	3 self-review rounds; critic rounds unavailable (none configured)
complete	—	—

Pull requests

PR	Scope / tasks	Status
#170	the whole work item — T1–T7	open

Progress entries

2026-08-06 — spec chain written and locked

• Phase: requirements-definition → tasks-breakdown
• Did: wrote requirements.md, design.md, testing-plan.md, tasks.md. Chose option 2 (validates:) with option 3 (fail closed) as its backstop, and rejected option 1 because produces means authorship and the manifest's deliberately phase-less execution-log.md entry is exactly what P1/P2 key on.
• Checkpoint/tests: none — no code yet.
• Next: T1, test-first.

2026-08-06 — implementation

• Phase: implementation
• Did: T1–T6. validate-artifacts resolves validates: through the shared resolve_produces and fails closed when content checks resolve no artifact; the six review nodes gate execution-log.md; the bundled template gained ## Capability docs; P5a/P5b/P5c landed; a new test_graph_review_chain_integration.py drives all six shipped nodes; decision-063 and the capability doc written.
• Checkpoint/tests: pytest -q → 1380 passed, 1 skipped. ruff, pyright clean.
• Next: T7 — execute the testing plan, commit evidence.

2026-08-06 — verification

• Phase: verification
• Did: T7. Ran every activity, committed evidence under evidence/. Confirmed P5a fails against the pre-fix graph naming all six nodes, and P5c fails against the pre-fix template naming Capability docs — the checks check something. Ran the-loop's own gate against this work item (the-loop check issue-167 --recompute --fail-on block, exit 0).
• Checkpoint/tests: testing-plan.md § Verification results.
• Next: self-review, then the PR briefing.

Review cycles

Outcome is one of: new findings · zero (converged) · escalated · unavailable (the configured critic could not run — it does NOT count toward reviews.criticReviewCount).

Cycle	Type (self/critic/security)	Reviewer	Outcome	Link
1	self	the-loop (agent)	new findings — (a) P3 read a node's sections: without asking which file they were meant for, so a node carrying both produces: and validates: would have demanded the log's sections of its own template; _required_sections now ignores entries that name a validates: target. (b) The optional: early return judged entry on all slots, so an optional node validating the shared log would have been treated as entered — the log exists for every work item; it is now judged on what the node authored.	this PR
2	self	the-loop (agent)	new findings — the ticket's reproduction script does not go quiet after the fix: it tests for sections: without produces:, a question that predates the validates: vocabulary. Recorded as it happened in evidence/reproduction.md, with the corrected structural check and the behavioural check beside it, rather than quietly swapping the activity.	evidence/reproduction.md
3	self	the-loop (agent)	new finding, accepted not fixed — a previously-completed work item re-checked with --recompute now blocks at capability-docs: issue-161, issue-163 and issue-165 all do, because the template never offered the section and their logs never carried it. That is the gate being honest about a missing record, and backfilling the section into merged logs would be fabricating one. CI is unaffected (the-loop-gate.yml gates only work items the PR touches). Flagged to the reviewer instead.	this PR
4	self	the-loop (agent)	zero (converged)	this PR
5	critic	—	unavailable — reviews.critics: []; no critic harness is configured in this repository, so no critic round could run. Does not count toward criticReviewCount; the human PR review is the backstop.	.the-loop/harness-config.yaml

Security review (gate)

Required before ready-to-ship (security.review.required). See reference/security.md.

• Mechanism: the-loop checklist, cross-checked against design.md § Security design (security.review.mechanism: auto).
• Outcome: pass. The change adds no external input, no network reach and no new privilege. validates: is read from the shipped graph, which a repository cannot define or override (_warn_on_repo_graph), so a work item cannot point a gate at a path of its choosing; every target resolves through resolve_produces onto the work item's own spec directory, and the graph declares bare filenames. Every new branch blocks — skip is returned in strictly fewer situations after this change than before. Findings carry the-loop's own vocabulary plus repo-relative paths (R3.6), unchanged. The one accepted limit is written down rather than implied: the section check is structural, so placeholder text passes it — the gate proves the record exists, and the reviewer judges whether the review was any good.
• Human sign-off: n/a — risk tier 3, below security.review.humanSignOffMinTier: 4.

Final validation evidence

Every acceptance criterion is proved by a committed artifact under evidence/; the per-activity record is in testing-plan.md § Verification results.

• The gates fire (R1, R2, R3) — all six review nodes now block where they used to skip, driven over the shipped graph: evidence/reproduction.md. 12 new hook unit tests and 24 integration assertions: evidence/tests.md.
• The assertions assert something (R4) — P5a fails naming all six nodes against the pre-fix graph; P5c fails naming Capability docs against the pre-fix template: evidence/tests.md.
• The template can clear its own gates (R3.3) — the bundled execution log, unedited, passes all six nodes (test_the_bundled_template_can_clear_every_gate_in_the_chain).
• Nothing else moved — 1380 passed, 1 skipped (pre-existing, unrelated); ruff, ruff format --check, pyright and markdownlint clean: evidence/tests.md, evidence/lint-and-types.md.
• the-loop's own gate passes on this work item — the-loop check issue-167 --recompute --fail-on block exits 0 (parked at requirements-approval, which is the human gate this PR is).

Capability docs

Which living capability docs this work item changed, and the history row that traces the behaviour back to it. Updated in the same PR as the change — a ready-to-ship gate item (workflow.capabilitiesDir).

Capability doc	What changed	History row
docs/capabilities/process-graph.md	new section What a node validates: the parameter, the shared resolver, the absent-target block, the fail-closed rule with its not-retriable block, P5, and the explicit statement of what a structural section check does and does not prove	issue-167 · decision-063

No other capability doc changed: the fix is internal to the process graph's own gating, and spec-workflow.md / testing-and-contracts.md describe artifacts whose shape is untouched. skills/the-loop/reference/workflow.md gained one clause pointing the capability-docs fold-in step at the log section that now gates it — a skill edit, not a capability doc.