test(ceo): cover ceo_open_guidance_pr + harden the redline branch (fixes #250 RC) #251

Merged

joshtronic merged 1 commit from fix/ceo-guidance-test into master

2026-06-25 19:50:48 +00:00

igor commented

2026-06-25 19:43:41 +00:00

Collaborator

Fixes forward the valid REQUEST_CHANGES the shadow review left on #250 (which got merged past the RC — exactly the failure mode the binding gate is meant to stop).

The blocker (now fixed): ceo_open_guidance_pr — the riskiest new path (base64 decode → conditional section-create → bullet append → re-encode → contents PUT → PR open) — was verified by nothing, and the live run never exercised it (no guidance emitted with no proposal history). So the core PR-opening mechanics had never actually run.

This PR:

Test ceo_open_guidance_pr (the reviewer's exact ask — mirrors the ceo_file_proposal POST_BODY pattern): mock _fj to return a known CEO.md, invoke it, assert the captured PUT body's decoded content — creates the ## Decision guidance header when absent, does not duplicate it when present, keeps existing entries, appends the - <week>: <guidance> bullet, PUTs with the file sha, branches ceo-guidance-*.
Test ceo_guidance_pr_open (closes the loop): a ceo-guidance-* head throttles; otherwise it proceeds.
Non-blocking nits, both addressed: branch is now ceo-guidance-<week>-<epoch> so a merged-but-undeleted same-week branch can't block a later redline on the new_branch PUT; and the append-only placement assumption (Decision guidance stays the last section) is now an explicit comment.

test-ceo (9 new checks) + check-sync green.

Closes #238 — the final piece of the autonomous-CEO arc (#246 digest · #247 proposals · #250 decision-guidance redlines · #248 the .agent/ceo.md→CEO.md move). With it landed and tested, #238's core is realized: an opt-in (mandate-file) CEO pass that gathers repo state, LLM-decides / harness-executes, and produces a board digest + greenlight-me proposals + decision-guidance redlines — never shipping code (propose + ratify, every write human-gated).

Two spec items landed differently, by design:

PR reviews are the separate shadow-review pass's job (do_review_tick), not the CEO's — that pass runs on the non-author tier to honor the supply-chain trust boundary, which the CEO (same model writing the digest) can't.
Daily light-steering was deferred — weekly is the cadence ("start tight"). The mandate self-refinement (Phase 3) is the steering that earned its place.

Say the word if you want either built as follow-up and I'll reopen/keep #238 for it.

Fixes forward the **valid `REQUEST_CHANGES`** the shadow review left on #250 (which got merged past the RC — exactly the failure mode the binding gate is meant to stop). **The blocker (now fixed):** `ceo_open_guidance_pr` — the riskiest new path (base64 decode → conditional section-create → bullet append → re-encode → contents PUT → PR open) — was verified by *nothing*, and the live run never exercised it (no guidance emitted with no proposal history). So the core PR-opening mechanics had never actually run. **This PR:** - **Test `ceo_open_guidance_pr`** (the reviewer's exact ask — mirrors the `ceo_file_proposal` `POST_BODY` pattern): mock `_fj` to return a known `CEO.md`, invoke it, assert the captured **PUT body's decoded content** — creates the `## Decision guidance` header when absent, does **not** duplicate it when present, keeps existing entries, appends the `- <week>: <guidance>` bullet, PUTs with the file sha, branches `ceo-guidance-*`. - **Test `ceo_guidance_pr_open`** (closes the loop): a `ceo-guidance-*` head throttles; otherwise it proceeds. - **Non-blocking nits, both addressed:** branch is now `ceo-guidance-<week>-<epoch>` so a merged-but-undeleted same-week branch can't block a later redline on the `new_branch` PUT; and the append-only placement assumption (Decision guidance stays the last section) is now an explicit comment. `test-ceo` (9 new checks) + `check-sync` green. --- **Closes #238** — the final piece of the autonomous-CEO arc (#246 digest · #247 proposals · #250 decision-guidance redlines · #248 the `.agent/ceo.md`→`CEO.md` move). With it landed and tested, #238's core is realized: an **opt-in** (mandate-file) CEO pass that gathers repo state, **LLM-decides / harness-executes**, and produces a board digest + greenlight-me proposals + decision-guidance redlines — **never shipping code** (propose + ratify, every write human-gated). Two spec items landed differently, by design: - **PR reviews** are the separate **shadow-review** pass's job (`do_review_tick`), not the CEO's — that pass runs on the **non-author tier** to honor the supply-chain trust boundary, which the CEO (same model writing the digest) can't. - **Daily light-steering** was deferred — weekly is the cadence ("start tight"). The mandate self-refinement (Phase 3) is the steering that earned its place. Say the word if you want either built as follow-up and I'll reopen/keep #238 for it.

igor added 1 commit

2026-06-25 19:43:41 +00:00

test(ceo): cover ceo_open_guidance_pr + harden the redline branch (review #250 )

Lint / check-sync (pull_request) Successful in 4s

Details

Lint / check-sync (push) Successful in 4s

Details

4bf7331674

#250's shadow review correctly RC'd: ceo_open_guidance_pr -- the riskiest new
path (decode -> section-create -> append -> re-encode -> PUT -> PR) -- was
verified by nothing, and the live run never fired it (no guidance emitted with
no proposal history). It merged anyway; this fixes it forward.

- test: mock _fj to return a known CEO.md, invoke ceo_open_guidance_pr, assert
  the captured PUT body's decoded content -- creates the '## Decision guidance'
  header when absent, doesn't duplicate it when present, keeps existing entries,
  appends the '- <week>: <guidance>' bullet, PUTs with the file sha, branches
  ceo-guidance-*. Plus a ceo_guidance_pr_open throttle test (ceo-guidance head
  -> throttles; otherwise proceeds). (Mirrors the ceo_file_proposal POST_BODY
  pattern the reviewer pointed at.)
- harden (non-blocking nits from the review): branch is now
  ceo-guidance-<week>-<epoch> so a merged-but-undeleted same-week branch can't
  block a later redline on the new_branch PUT; and the append-only placement
  assumption (Decision guidance stays the last CEO.md section) is now an
  explicit comment.

test-ceo + check-sync green.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011KXPafSYJPY6XkWHHk1Lfs

igor requested review from joshtronic

2026-06-25 19:43:42 +00:00

igor commented

2026-06-25 19:46:07 +00:00

Author

Collaborator

🤖 Review — `APPROVE` (automated)

CI for 4bf73316: success

Summary: Small, test-focused follow-up that adds the missing coverage for ceo_open_guidance_pr / ceo_guidance_pr_open and hardens the redline branch name with an epoch suffix. CI is green, scope is tight (~50 lines, two files), and the description matches the diff.

Checklist verification (all honest):

"Test ceo_open_guidance_pr" — present in bin/test-ceo.sh; asserts on the decoded PUT body (header creation, no-duplicate, existing-entry retention, bullet append, sha, branch prefix). ✓
"Test ceo_guidance_pr_open" — present; covers both throttle and proceed cases. ✓
"branch is now ceo-guidance-<week>-<epoch>" — lib/ceo.sh:295 changed to ceo-guidance-${week}-$(date +%s). ✓
"append-only placement comment" — present at lib/ceo.sh:288. ✓

Correctness notes:

The tests are non-vacuous: an early return 1 would leave PUT_BODY empty and fail the has/eq assertions, so a green CI confirms the full decode→section-create→append→re-encode→PUT path actually ran.
Assertions are deliberately week-independent (bullet check is on ": Favor growth-lever work.", branch check is a prefix match), so they won't flake across a week boundary. Good.
The epoch-suffixed branch still matches the throttle's ceo-guidance-* prefix (test uses ceo-guidance-2026-W26-1), so the hardening doesn't break throttling.
No CI-config changes, no security surface touched, no secrets.

Minor observations (non-blocking):

The lib/ceo.sh change is purely a comment + branch-suffix tweak on the shown lines; the rest of the function body isn't in the diff, but it isn't being modified and the new tests exercise it end-to-end, so the green run is sufficient confirmation.
ceo_open_guidance_pr is invoked with >/dev/null 2>&1 || true, which swallows the exit code — acceptable here since correctness is asserted on the captured PUT body rather than the return value.

Clean, in-scope, and well-tested. No reservations.

_{Independent review by the harness on claude-opus-4-8. The human reviewer is requested once Igor has reviewed; a human still merges.}

### 🤖 Review — `APPROVE` _(automated)_ CI for `4bf73316`: **success** **Summary:** Small, test-focused follow-up that adds the missing coverage for `ceo_open_guidance_pr` / `ceo_guidance_pr_open` and hardens the redline branch name with an epoch suffix. CI is green, scope is tight (~50 lines, two files), and the description matches the diff. **Checklist verification (all honest):** - "Test `ceo_open_guidance_pr`" — present in `bin/test-ceo.sh`; asserts on the decoded PUT body (header creation, no-duplicate, existing-entry retention, bullet append, sha, branch prefix). ✓ - "Test `ceo_guidance_pr_open`" — present; covers both throttle and proceed cases. ✓ - "branch is now `ceo-guidance-<week>-<epoch>`" — `lib/ceo.sh:295` changed to `ceo-guidance-${week}-$(date +%s)`. ✓ - "append-only placement comment" — present at `lib/ceo.sh:288`. ✓ **Correctness notes:** - The tests are non-vacuous: an early `return 1` would leave `PUT_BODY` empty and fail the `has`/`eq` assertions, so a green CI confirms the full decode→section-create→append→re-encode→PUT path actually ran. - Assertions are deliberately week-independent (bullet check is on `": Favor growth-lever work."`, branch check is a prefix match), so they won't flake across a week boundary. Good. - The epoch-suffixed branch still matches the throttle's `ceo-guidance-*` prefix (test uses `ceo-guidance-2026-W26-1`), so the hardening doesn't break throttling. - No CI-config changes, no security surface touched, no secrets. **Minor observations (non-blocking):** - The `lib/ceo.sh` change is purely a comment + branch-suffix tweak on the shown lines; the rest of the function body isn't in the diff, but it isn't being modified and the new tests exercise it end-to-end, so the green run is sufficient confirmation. - `ceo_open_guidance_pr` is invoked with `>/dev/null 2>&1 || true`, which swallows the exit code — acceptable here since correctness is asserted on the captured PUT body rather than the return value. Clean, in-scope, and well-tested. No reservations. --- <sub>Independent review by the harness on `claude-opus-4-8`. The human reviewer is requested once Igor has reviewed; a human still merges.</sub>

igor added spent time

2026-06-25 19:46:07 +00:00

1 minute 17 seconds