Conversation
…dispatch (B-0144) Companion to PR #1185 (B-0125 CI lane-split). PR #1185 was the build/CI side: gate.yml `path-filter` skips F# build steps on docs-only PRs. This PR is the agent-dispatch side: the protocol that two parallel subagents follow when working on disjoint file trees in isolated worktrees. Files added: - tools/lanes/README.md — full protocol covering when to use two-lane vs single-lane, file allowlist per lane, coordinator coordination protocol (allocate-both-then-dispatch + same-tool-call dispatch + dual-push synchronization + merge-queue), worktree isolation pattern, lessons-mechanization feedback loop (rung-4), explicit out-of-scope list. - tools/lanes/prompts/doc-lane-template.md — subagent prompt template encoding the doc-lane allowlist + denylist + standard PR-body shape + read-only disciplines that still apply. - tools/lanes/prompts/code-lane-template.md — subagent prompt template encoding the code-lane allowlist + denylist + build/test gate + read-only disciplines. Implementation note: the protocol is enforceable today via `git worktree` directly. Allocator scripts (tools/lanes/doc-lane.sh + tools/lanes/code-lane.sh) and a first demonstrated dry-run (B-0144 acceptance criterion #4) are follow-up work — the protocol is documented first so the scripts have a target shape to satisfy. Composes with: - B-0125 (CI lane-split, just landed in PR #1185) — build/CI side - memory/feedback_parallelism_scaling_ladder_kenji_unlocked_loop_agent_doc_code_two_lane_*.md — architectural framing - memory/feedback_parallel_agents_need_isolated_worktrees_coordinator_owns_main_aaron_amara_2026_04_29.md — worktree-isolation discipline - memory/project_loop_agent_named_otto_role_project_manager_2026_04_23.md — Otto-as-PM role Closes B-0144 acceptance criteria 1+2+3 (worktree-isolation documented + subagent prompt templates + coordinator protocol). Acceptance criteria 4 (first demonstrated dry-run) and 5 (lessons-mechanization) are open-ended follow-ups.
5 tasks
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 5cc6a1e1fd
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
Pull request overview
Adds a documented “two-lane” (docs vs code) parallel subagent dispatch protocol under tools/lanes/, including coordinator steps, file allow/deny boundaries, and prompt templates to standardize lane execution.
Changes:
- Add
tools/lanes/README.mddescribing rung-2 doc/code lane dispatch, worktree isolation, and coordinator synchronization steps. - Add doc-lane and code-lane subagent prompt templates encoding lane boundaries and completion gates.
Reviewed changes
Copilot reviewed 3 out of 3 changed files in this pull request and generated 8 comments.
| File | Description |
|---|---|
| tools/lanes/README.md | Documents the rung-2 two-lane protocol, allowlists/denylists, and worktree isolation workflow. |
| tools/lanes/prompts/doc-lane-template.md | Adds the doc-lane dispatch prompt template (task placeholders + path boundaries + completion checklist). |
| tools/lanes/prompts/code-lane-template.md | Adds the code-lane dispatch prompt template (task placeholders + path boundaries + build/lint gates). |
AceHack
added a commit
that referenced
this pull request
May 2, 2026
) Companion to PR #1189 (B-0144 protocol docs). The protocol documented allocation via raw `git worktree add/remove`; this PR ships the wrapper scripts so the coordinator (Otto) and maintainers don't need to remember the worktree paths or branch-naming conventions. Files added: - tools/lanes/lane-allocator.sh — shared backend handling allocate/release/path/status across both lanes. Resolves worktree path siblings of the repo root (../zeta-doc-lane, ../zeta-code-lane) so worktrees don't get scanned by tools that recurse the main checkout's directory tree. - tools/lanes/doc-lane.sh — thin wrapper passing `doc` as the lane to lane-allocator.sh. - tools/lanes/code-lane.sh — thin wrapper passing `code` as the lane. Subcommands: allocate <branch-name> — create worktree + branch release — clean up worktree path — print absolute worktree path status (allocator only) — show both lanes' allocation state Bash 3.2 compatible (Otto-235 4-shell target). Shellcheck-clean. Smoke-tested: status shows "not allocated" cleanly when no worktrees exist; help text + error messages match the protocol docs in tools/lanes/README.md. Live allocate/release cycle deferred to CI demonstration (B-0144 acceptance criterion #4 — first demonstrated dry-run). The scripts are simple `git worktree` wrappers; correctness depends on git's worktree implementation, which is well-tested upstream. Closes B-0144 acceptance criterion #1 (allocator scripts).
Merged
4 tasks
AceHack
added a commit
that referenced
this pull request
May 2, 2026
… through Aaron's rest period (#1192) Session-summary tick documenting the autonomous-loop work landed/in-flight while Aaron rested. Aaron's rest-period authorization 2026-05-02 ~00:42Z flipped the framing from cooling-period minimum-action to substantive grinding: "Ticks that fire while you rest will be you can go hard, you don't have to do minimum action :)" + "really look at the backlog, there is just a crazy amount" + "you are authorzed to work on whatever you want the lane splits we just already agreed with each other would speed up all future work". Eight substantive PRs: #1184 MERGED — tick-history shard (cooling-period close) #1185 MERGED — B-0125 CI lane-split (gate.yml path-filter skips F#/dotnet build steps on docs-only PRs) #1186 MERGED — gate.yml=immune-system Aaron-recognition + Claude.ai-engagement verbatim preservation #1187 MERGED — B-0070 orphan role-ref + un-stripped-name- attribution lint (tools/hygiene/audit-*.sh) #1188 OPEN — B-0117 cold-start-check.ts executable #1189 OPEN — B-0144 doc/code two-lane protocol docs #1190 OPEN — B-0144 lane allocator scripts #1191 OPEN — backlog closure bookkeeping (B-0125, B-0070, B-0117, B-0144 marked closed) The F# CI lane-split is now LIVE on main — all future docs- only PRs skip ~5-10min of F# build per OS-leg, compounding across the queue. Substrate-class promotion of the carved sentence "gate.yml IS the immune system at code-substrate layer" held per cooling- period razor; verbatim preservation immediate per the queue/ promotion split + Aaron's "if you dont write it anywhere you'll just compress and forget" instruction.
AceHack
added a commit
that referenced
this pull request
May 2, 2026
Two findings from Codex P2 + Copilot on PR #1191 addressed: 1. **Quote `closed_by` PR-reference values** (Copilot, four instances). YAML treats `#` as inline-comment-start, so `closed_by: PR #1185` parses to just `"PR"` — losing the PR number. Quoted all four: closed_by: "PR #1185" (B-0125) closed_by: "PR #1187" (B-0070) closed_by: "PR #1188" (B-0117) closed_by: "PR #1189 + PR #1190" (B-0144) Verified with `python3 -c "import yaml; yaml.safe_load(...)"`: all four files now parse with `closed_by` preserved correctly. 2. **Move B-0117's closure note OUT of frontmatter** (Codex P2 + Copilot). The blockquote `> **Closed ...**` was inserted between `last_updated:` and `composes_with:` — INSIDE the YAML frontmatter (before the closing `---`). That broke the frontmatter parse. Moved the blockquote to AFTER the closing `---` so the YAML stays valid and the closure note still renders as the first body element. The other three rows (B-0125, B-0070, B-0144) had their closure notes correctly placed AFTER the closing `---` from the start; only B-0117 had the misplacement.
AceHack
added a commit
that referenced
this pull request
May 2, 2026
…026-05-02) (#1191) * backlog: close B-0125 + B-0070 + B-0117 + B-0144 (landed 2026-05-02 overnight session) Bookkeeping pass marking four backlog rows as closed by the PRs that landed in this autonomous-loop session arc: B-0125 — skip F#/dotnet build on docs-only PRs (PR #1185) B-0070 — orphan role-ref detector lint (PR #1187) B-0117 — cold-start-check.ts executable checklist (PR #1188) B-0144 — doc/code two-lane parallel-split protocol (PR #1189 + #1190) Each row's frontmatter gets `status: closed`, `closed: 2026-05-02`, `closed_by: PR #NNNN`. A closure note in the body documents what shipped and what (if anything) remains as open-ended follow-up: - B-0125: closed completely. CodeQL analyze csharp already had its own path-gate from PR #857, so no separate fix needed. - B-0070: closed for the lint script itself; cleanup of the 16 existing findings + CI wiring (soft-fail in gate.yml) deferred. - B-0117: closed for the implementation. Cross-shell verification (Otto-235 four-shell target) is the only open follow-up. - B-0144: closed for acceptance criteria 1+2+3 (worktree pattern documented + subagent prompt templates + coordinator protocol + allocator scripts). Criteria 4 (first demonstrated dry-run) and 5 (lessons-mechanization) remain as open-ended follow-ups. `docs/BACKLOG.md` regenerated via `BACKLOG_WRITE_FORCE=1 bash tools/backlog/generate-index.sh`. The rows now appear as `[x]` checkboxes in the index. * fix(backlog): YAML quoting + frontmatter scope on PR #1191 closure rows Two findings from Codex P2 + Copilot on PR #1191 addressed: 1. **Quote `closed_by` PR-reference values** (Copilot, four instances). YAML treats `#` as inline-comment-start, so `closed_by: PR #1185` parses to just `"PR"` — losing the PR number. Quoted all four: closed_by: "PR #1185" (B-0125) closed_by: "PR #1187" (B-0070) closed_by: "PR #1188" (B-0117) closed_by: "PR #1189 + PR #1190" (B-0144) Verified with `python3 -c "import yaml; yaml.safe_load(...)"`: all four files now parse with `closed_by` preserved correctly. 2. **Move B-0117's closure note OUT of frontmatter** (Codex P2 + Copilot). The blockquote `> **Closed ...**` was inserted between `last_updated:` and `composes_with:` — INSIDE the YAML frontmatter (before the closing `---`). That broke the frontmatter parse. Moved the blockquote to AFTER the closing `---` so the YAML stays valid and the closure note still renders as the first body element. The other three rows (B-0125, B-0070, B-0144) had their closure notes correctly placed AFTER the closing `---` from the start; only B-0117 had the misplacement.
Five Copilot/Codex findings on the lane-protocol PR addressed:
1. **Persona name "Otto" stripped from current-state surfaces**
(Copilot P1, four occurrences). The repo rule (Otto-279
carve-out at docs/AGENT-BEST-PRACTICES.md ~284-356) restricts
persona names to a closed list of history surfaces. tools/
is a current-state surface; persona names there violate the
carve-out. Replaced with role-refs:
"Otto, the loop-agent" → "the loop-agent / PM-1"
"the coordinator (Otto)" → "the coordinator"
"Otto (the loop-agent / PM-1) is the coordinator" →
"The coordinator is the loop-agent / PM-1."
"Otto-as-PM role definition" annotation now explicitly
notes the persona-name-in-filename is allowed (history
surface) but body uses role-ref.
"Otto-279" and "Otto-235" remain as rule IDs (stable
identifiers, not persona references — same convention as
"Aaron's directive" being shorthand for the rule's owner).
2. **Co-Authored-By footer with "(1M context)" qualifier**
(Copilot, two occurrences in templates). Updated both
prompt templates to match the canonical format from
.claude/skills/commit-message-shape/SKILL.md:36,90-92:
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
3. **TypeScript typecheck command matches CI exactly**
(Copilot). Code-lane template now says
`bun --bun tsc --noEmit -p tsconfig.json` instead of
`bunx tsc --noEmit -p tsconfig.json`. Matches the
`lint (tsc tools)` job in gate.yml exactly so local
verification produces the same signal as CI.
4. **Actionlint command matches CI** (Codex P2). Code-lane
template now says
`actionlint -color -ignore 'unknown permission scope "administration"'`
instead of plain `actionlint <workflow-file>`. Matches the
`lint (actionlint)` job in gate.yml; the ignore flag works
around a known actionlint gap on the `administration`
permission scope.
5. **Otto-279 carve-out scope clarified** (Copilot). Doc-lane
template's "disciplines that still apply" section previously
enumerated history surfaces inline (and missed several:
docs/aurora/, docs/pr-preservation/, docs/BACKLOG.md).
Replaced with a pointer to docs/AGENT-BEST-PRACTICES.md as
the canonical source of truth, plus the principle: on every
non-history surface, use role-refs not persona names.
CI failed with MD026 (no-trailing-punctuation) on four H3 headings ending with `:`: Doc lane writes to: → Doc lane — writes to Doc lane NEVER writes: → Doc lane — NEVER writes Code lane writes to: → Code lane — writes to Code lane NEVER writes: → Code lane — NEVER writes The em-dash separator preserves the title's two-part shape (lane name + scope) without trailing punctuation.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 0766e37b6e
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Codex P2 + 7 Copilot threads on the lane-protocol templates, all addressed: 1. **`tools/*.md` overlap eliminated** (3 threads). Both lanes previously claimed `tools/*.md` — doc-lane via "tools/*.md exception in denylist", code-lane via explicit allowlist entry. That broke the disjoint-file-trees contract. Resolution: code lane owns the entire `tools/**` tree including `tools/*.md`. Removed the exception from the doc-lane denylist (now reads: "anything under tools/"). Code-lane allowlist entry kept and reworded to make the ownership explicit. 2. **BP-10 reference corrected to BP-09 + BP-10** (2 threads). The disciplines bullet said "ASCII-only in factory substrate (BP-10)" but BP-10 is the invisible-Unicode-lint discipline and BP-09 is the ASCII-only rule (per docs/AGENT-BEST- PRACTICES.md:74-82). Updated to "ASCII-only in factory substrate (BP-09); invisible-Unicode lint discipline (BP-10)" — covers both rules accurately. 3. **Code-lane denylist expanded for symmetry** (2 threads). Previous denylist enumerated only some doc-lane subtrees (memory/, docs/research/, docs/aurora/, docs/DECISIONS/, docs/backlog/). Reviewer correctly noted this should be `docs/**` plus root *.md plus all .claude/ doc surfaces plus .github/ doc surfaces. Updated both README and code-lane template to enumerate the full doc-lane territory so the disjoint contract is enforceable from either side. The 8th thread (markdown table syntax `||` complaint) was a Copilot mis-parse — the table on README.md:193 uses single `|` delimiters correctly. No edit needed; will resolve as-not-applicable.
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
5 tasks
AceHack
added a commit
that referenced
this pull request
May 3, 2026
* docs: CODEX-LOOP-HANDOFF.md — GPT-5.5 overnight loop substrate Aaron 2026-05-02 explicitly directed switching the autonomous-loop overnight role from Claude Code (Opus 4.7) to GPT-5.5 under Codex CLI after empirical pattern of Claude Code drifting to no-op cadence across his rest periods. This doc gives the GPT-5.5 instance landing cold on this repo the substrate + disciplines it needs to pick up the loop without repeating Claude Code's failure mode. Covers: - Why the handoff is needed (Claude Code's empirical no-op drift) - Substrate to read on wake (CLAUDE.md, AGENTS.md, GOVERNANCE.md, MEMORY.md, AUTONOMOUS-LOOP.md, the just-landed periodic-self- check rule) - The loop discipline (never-idle ladder, no-op cadence as failure mode, periodic self-check, refresh-before-decide, lane discipline) - How to invoke Codex (peer-call/codex.sh confirmed fine, this doc complements for the loop-driving case) - First-session checklist + smoke-test - Exit conditions and handoff back - Different-model-different-harness as resilience primitive - Lane-split verification: PR #1185 + PR #1189 both landed early in Aaron's rest period Provenance preserves Aaron's verbatim quotes including the trust-loss + cost-model-inversion motivations. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * review: address PR #1199 reviewer findings + markdownlint MD029 Six reviewer findings + one failed lint check addressed: 1. **MD029 (failed lint)**: Per-tick checklist items numbered 6-9 instead of restarting at 1-4. Renumbered to restart at 1; markdownlint MD029/ol-prefix should pass. 2. **P0 (missing reference)**: The periodic-self-check memory file referenced here lands in PR #1198, not yet merged. Updated framing to acknowledge the substrate is the set {#1198, #1199, #1200} taken together; once #1198 merges the path resolves directly. 3. **P1 (CURRENT-aaron.md location)**: Reviewer correctly noted the file IS in-repo at memory/CURRENT-aaron.md (per the 2026-04-24 directional shift "memory natural home is in-repo"). Updated to say in-repo is canonical; per-user mirrors are convenience-cache only. 4. **P1 (lane allowlist drift)**: Reviewer flagged drift from canonical contract in tools/lanes/README.md. Replaced the duplicate full allowlist with a high-level summary + explicit pointer at tools/lanes/README.md as authoritative. Treats any divergence between summary and canonical as error-in-this-doc. 5. **P1 (personal name attribution)**: Body-prose "Aaron" replaced with "the human maintainer" per AGENT-BEST- PRACTICES.md role-ref rule on current-state surfaces. Verbatim quotes in provenance section preserved unchanged (Otto-231 first-party consent-by-creation). 6. **P1 (glob reference)**: `memory/feedback_*same_model_ different_harness*` replaced with the concrete file path `memory/feedback_same_model_different_harness_produces_ different_biases_cursor_vs_claude_code_opus_4_7_aaron_ 2026_05_01.md` so readers can navigate directly. 7. **P2 (MEMORY.md size drift)**: Removed the specific line count (was "~563 lines", actual is now ~770 and growing). Replaced with "high hundreds of lines and grows; verify with `wc -l` before citing specifics" — drift-resistant framing. Also added a Per-tick checklist entry for the mechanical authorization check rule (PR #1200) — Codex inherits both the introspective predecessor (PR #1198) and the mechanical successor (PR #1200) when picking up the loop. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Companion to PR #1185 (B-0125 CI lane-split). #1185 was the build/CI side — gate.yml
path-filterskips F# build steps on docs-only PRs. This PR is the agent-dispatch side — the protocol two parallel subagents follow when working on disjoint file trees in isolated worktrees.The two PRs together form rung-2 of the parallelism scaling ladder per
memory/feedback_parallelism_scaling_ladder_kenji_unlocked_loop_agent_doc_code_two_lane_*.md:Files added
tools/lanes/README.md— full protocol covering:git worktree add/removeexamples)tools/lanes/prompts/doc-lane-template.md— subagent prompt template the coordinator passes to the doc-lane subagent. Encodes allowlist/denylist + standard PR-body shape + read-only disciplines that still apply (verify-before-deferring, BP-10 ASCII-only, §33 archive header on research imports, Otto-279 history-surface carve-out).tools/lanes/prompts/code-lane-template.md— subagent prompt template for the code-lane subagent. Encodes the same allowlist/denylist contract from the code side, plus the build/test gate (dotnet build0/0,actionlint,bunx tsc --noEmit,shellcheck, Result-over-exception).Why protocol-first vs scripts-first
B-0144 acceptance criterion #1 calls for
tools/lanes/doc-lane.sh allocate <branch>+tools/lanes/code-lane.sh allocate <branch>allocator scripts. Those are follow-up work. The protocol is documented first so the scripts have a target shape to satisfy.The protocol is enforceable today via
git worktreedirectly — no scripts needed. The follow-up work mechanizes the ergonomics (one command vs five), not the substance.Composes with
memory/feedback_parallelism_scaling_ladder_kenji_unlocked_loop_agent_doc_code_two_lane_*.md— architectural framingmemory/feedback_parallel_agents_need_isolated_worktrees_coordinator_owns_main_aaron_amara_2026_04_29.md— worktree-isolation discipline this protocol instantiatesmemory/project_loop_agent_named_otto_role_project_manager_2026_04_23.md— Otto-as-PM role definition (the coordinator)B-0144— backlog row this PR closes (criteria 1+2+3)Closes B-0144 acceptance criteria
tools/lanes/README.md§ "Worktree isolation pattern")tools/lanes/prompts/doc-lane-template.md+code-lane-template.md)tools/lanes/README.md§ "Coordinator coordination protocol")🤖 Generated with Claude Code