Execution Log: wire the ingress to the process graph

Phase transitions

Phase	Entered	Reviewed/approved by	Notes
requirements-definition	2026-07-29	@MadaraUchiha-314 (PR #114: "approved")	No brainstorm — gap established by code tracing, recorded on the ticket
design	2026-07-29	@MadaraUchiha-314 (PR #114: "approved")
tasks-breakdown	2026-07-29	@MadaraUchiha-314 (PR #114: "approved")
implementation	2026-07-29	@MadaraUchiha-314 (PR #114: "approved")	T1–T10
needs-review	2026-07-29	@MadaraUchiha-314 (PR #114: "approved")	Tier-4 gates cleared by that comment; completes when #114 merges
complete

Pull requests

PR	Scope / tasks	Status
#114	spec + T1–T9	open

Progress entries

2026-07-29 — spec written (requirements → design → tasks)

• Phase: tasks-breakdown → implementation
• Did: Traced the gap in the tree (graph runtime has one importer; HookContext.event has zero writers; no node is ever entered on the automated path). Opened #113. Wrote requirements.md (13 EARS ACs + threat-model-lite, risk tier 4), design.md (GraphLink seam in the shared dispatcher + Runtime.start()), tasks.md (8-task DAG).
• Checkpoint/tests: none yet — no code written.
• Next: T1 — Runtime.start(), red first.
• Blockers: none. Tier 4 means human-approves-pr and a named human security sign-off before completion; both are requested on the PR.

2026-07-29 — implementation complete (T1–T9)

• Phase: implementation → needs-review
• Did: Runtime.start() + graph.started; graphlink.py (spec_id_for, comments_from, GraphLink with all five skip paths); advance(..., event=) so HookContext.event finally has a writer; graph/bootstrap.build_runtime() shared with graph_cmd; routing.graph config block (schema + template); two dispatcher call sites, rebuilt on hot reload.
• Unplanned, in scope (T9): the first integration test to advance a real approval node parked with no edge ... on 'pass'. ChainOutcome.outcome read the routing value only from a blocking result, so classify-feedback's verdict — returned on a passing result — was discarded, and all three human-approval nodes in pdlc.yaml could never route. Pre-existing since issue-109, never caught because every existing test calls the hook directly rather than through advance(). Fixed; AC6 depends on it.
• Checkpoint/tests: make test → 749 passed, 1 skipped (was 741 before this work item; +29 new). ruff check clean, ruff format clean, pyright 0 errors, markdownlint 0 errors, validate_config.py all VALID. Red→green recorded per task: T1 AttributeError: no attribute 'start' → pass; T2–T5 ModuleNotFoundError → pass; T6 no attribute 'graphlink' → pass; T9 assert 'pass' == 'approved' → pass.
• Next: human review of the PR (tier 4 gate) + named security sign-off.
• Blockers: the two tier-4 gates below, both awaiting a human.

2026-07-29 — CI found a cross-repo collision (T10)

• Phase: needs-review
• Did: The gate job reported three work items instead of one — running the suite had written graph-state.json into the real docs/specs/issue-1 and docs/specs/issue-15, and appended an entry to issue-15's execution log. Not test noise: the dispatcher tests use github:octo/repo#15 → issue-15 with session cwd ., which is exactly the production shape under the default spawnWorkdir: ".". The link now requires the checkout's origin remote to name the work item's own <owner>/<repo>, failing closed when it cannot be read (AC14/A6). Reverted the three polluted files.
• Checkpoint/tests: four red tests first (mismatched origin, no origin, not a checkout, matching origin) → green. Full suite 758 passed, 1 skipped; running it now leaves git status clean, which is the real regression check.
• Next: the two tier-4 human gates.
• Blockers: unchanged — human PR approval and named security sign-off.

Review cycles

Cycle	Type (self/critic/security)	Reviewer	Outcome	Link
1	CI gate (the-loop's own)	the-loop gate job	Found the cross-repo collision (A6) — fixed in T10	run
2	human (PR approval + security sign-off)	@MadaraUchiha-314	approved	comment

Security review (gate)

• Mechanism: the-loop checklist (security.review.mechanism: auto)
• Outcome: pass. Reviewed against the threat model in requirements.md. The new boundary (untrusted comment text → hook chain) is enforced by classify-feedback's existing authorizedUsers + is_self_authored filter, which the link deliberately does not duplicate. A6 (cross-repository spec collision) was found by CI during review and closed by the origin-remote check in T10; every abuse case A1–A6 has a negative test. No new dependency, no new credential surface.
• Human sign-off: @MadaraUchiha-314 — PR #114, comment "approved" (2026-07-29). Required because riskTier 4 ≥ security.review.humanSignOffMinTier: 4. That comment replied to a PR body naming both open tier-4 gates (human-approves-pr and the security sign-off), and is recorded as clearing both.

Final validation evidence

• Test suite: 758 passed, 1 skipped (741 before this work item; +32 tests).
• CI on 7982db3: checks green. gate red only on the then-unlocked artifacts.
• The-loop's own gate, after locking: the-loop check issue-113 --recompute --fail-on block → exit 0, work item at requirements-approval in WAIT ("no authorized feedback yet"), which is the correct state for an open PR.
• AC coverage: AC1–AC3 test_graph_runtime.py::test_start_*; AC4/AC9/AC11/AC12 + A6/AC14 test_graphlink.py; AC5/AC6/AC8 test_graphlink.py::test_on_event_*, test_spec_id_for_*; AC6/AC10 + A1 test_graphlink_integration.py; AC7 by advance()'s existing wait/block paths; AC13 test_eventlog.py's catalog drift test.
• Regression check for A6: running the full suite leaves git status clean — the defect's signature was the suite writing into the repo's own docs/specs.

The feature, demonstrated by this very comment

The owner's approved on PR #114 is exactly the input this work item makes reachable: before it, HookContext.event had no writer, so classify-feedback returned waiting("no authorized feedback yet") no matter what any reviewer wrote — the state --recompute still shows here, because check is pure and passes no event. With a running daemon, that same comment now arrives as an issue_comment event, the dispatcher advances the graph with it attached, and the gate resolves.