[agento] [antig] feat(merge-train): add domain lock registry and hook configs#641
[agento] [antig] feat(merge-train): add domain lock registry and hook configs#641jleechan2015 wants to merge 1 commit into
Conversation
📝 WalkthroughWalkthroughA new ChangesAutonomous Harness Package and Integration
Estimated code review effort🎯 4 (Complex) | ⏱️ ~45 minutes Suggested reviewers
Poem
🚥 Pre-merge checks | ✅ 2 | ❌ 3❌ Failed checks (2 warnings, 1 inconclusive)
✅ Passed checks (2 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Comment |
![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: 2ff3f31230
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".
| const result = await spawnAOWorker({ | ||
| model, | ||
| systemPrompt: getSystemPromptForPhase(nextPhase), | ||
| taskPrompt, | ||
| workspace: projectPath, | ||
| sessionName, | ||
| skillRoot: opts.skillRoot, | ||
| runtime: opts.runtime, | ||
| projectId: opts.projectId, | ||
| }); |
There was a problem hiding this comment.
When the next phase is pre-spawned here, its fresh AO worktree never receives the untracked harness_state.json or prior sprint_* artifacts; only the non-standby path copies those files later. With the worktree/clone workspace plugins creating clean Git workspaces, the standby prompt and pollUntilActive() both wait on a file that does not exist in the standby worktree, so every completed phase waits the full maxIterationsPerPhase before falling back and leaves the standby session running.
Useful? React with 👍 / 👎.
| MATCH=$(echo "$CMD" | grep -oE '\-\-body[[:space:]]+'"'"'[^'"'"']*'"'"'([^[:space:]]+)?' 2>/dev/null || true) | ||
| if [ -n "$MATCH" ]; then | ||
| BODY=$(echo "$MATCH" | sed "s/--body[[:space:]]*'"'"'//; s/'"'"'$//" | tr '\n' ' ' | sed 's/[[:space:]]*$//') | ||
| fi |
There was a problem hiding this comment.
With set -u, BODY is now only assigned inside the single-quoted match branch; when the command uses a double-quoted body, --body-file, stdin, or no body, the next [ -z "$BODY" ] expansion aborts the hook with BODY: unbound variable before the fallback can run. This blocks valid gh pr create/edit --body "..." invocations with a hook crash instead of validating the evidence.
Useful? React with 👍 / 👎.
| // annotation and eval are orchestration phases — use orchestratorModel. | ||
| const model = (phase === "eval" || phase === "annotation") | ||
| ? orchestratorModel | ||
| : generatorModel; |
There was a problem hiding this comment.
The CLI exposes --evaluator-model, but eval workers are selected with orchestratorModel here, and evaluatorModel is otherwise only stored in state. In runs where a user supplies a distinct evaluator model, evaluation still launches with the orchestrator model, making the option ineffective for the phase it is documented to control.
Useful? React with 👍 / 👎.
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 5
🧹 Nitpick comments (1)
.claude/hooks/claim-verifier.sh (1)
73-73: 💤 Low valueConsider bash parameter expansion instead of sed.
Shellcheck suggests using
${variable//search/replace}for simpler string manipulation, which would avoid the complex quote escaping in the current sed command.Example refactor
# Current approach with sed: BODY=$(echo "$MATCH" | sed "s/--body[[:space:]]*'"'"'//; s/'"'"'$//" | tr '\n' ' ' | sed 's/[[:space:]]*$//') # Alternative with parameter expansion: BODY="${MATCH#--body*\'}" # Remove prefix through first ' BODY="${BODY%\'*}" # Remove suffix from last ' BODY=$(echo "$BODY" | tr '\n' ' ' | sed 's/[[:space:]]*$//')🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In @.claude/hooks/claim-verifier.sh at line 73, Replace the complex sed-based extraction of BODY from MATCH with bash parameter expansion: strip the leading "--body" prefix from MATCH using ${MATCH#...} to remove up to the opening single-quote, then strip the trailing single-quoted suffix using ${VAR%...} to remove from the last single-quote, assign the result to BODY, and then normalize newlines and trim trailing whitespace (e.g., via tr and a final sed/parameter expansion). Update the code around the BODY and MATCH handling in claim-verifier.sh to use these expansions instead of the long sed expression.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In @.claude/hooks/claim-verifier.sh:
- Around line 70-82: The script uses BODY in conditional checks before it is
guaranteed to exist so an externally set BODY can short-circuit the fallback;
initialize BODY to an empty string before performing the grep/extraction (i.e.,
add a BODY="" declaration before the first MATCH/grep block that references CMD
and MATCH) so the single-quoted extraction and the subsequent [ -z "$BODY" ]
check behave deterministically when running claim-verifier.sh.
In `@autonomous-harness/package.json`:
- Around line 7-17: The package's root export currently points to the CLI
(exports "." -> "dist/cli.js") so consumers importing the package get the CLI
instead of the library API defined in src/index.ts; update package.json so the
root export "." maps to the library entry (set "types" to "dist/index.d.ts" and
"import"/"require" to "dist/index.js" or adjust "main" to "dist/index.js"),
while keeping the "./cli" export pointing to "dist/cli.js"; modify the "exports"
block and optionally change "main" from "dist/cli.js" to the library entry to
ensure consumers receive the public API from src/index.ts after build.
In `@autonomous-harness/src/cli.ts`:
- Around line 17-19: Replace the rest-argument signatures on the CommanderLike
interface with explicit named parameters: change requiredOption(name: string,
...args: unknown[]) to something like requiredOption(name: string, description?:
string): CommanderLike; change option(name: string, ...args: unknown[]) to an
explicit signature such as option(name: string, flags?: string, description?:
string, defaultValue?: unknown): CommanderLike; and replace action(handler:
(...args: unknown[]) => unknown) with a single explicit parameter type (for
example action(handler: (args: unknown[]) => unknown): CommanderLike) or another
concrete parameter list that matches how the handler is invoked; update the
declarations for requiredOption, option, and action accordingly so there are no
rest parameters.
In `@autonomous-harness/src/orchestrator.ts`:
- Around line 95-98: The model-selection logic currently sets model based on
nextPhase === "eval" || "annotation" and always prefers opts.orchestratorModel,
so opts.evaluatorModel is ignored; update the conditional used where model is
assigned (the expression that reads nextPhase and opts.orchestratorModel /
opts.generatorModel) to specifically use opts.evaluatorModel when nextPhase ===
"eval" (e.g., for eval: opts.evaluatorModel ?? opts.orchestratorModel ??
"minimax/MiniMax-M2.7"), keep annotation using opts.orchestratorModel, and apply
the same fix to the second identical occurrence later in the file (the other
model selection at lines 555–557) so eval actually honors --evaluator-model.
- Around line 727-735: The standby branch persists the standby state via
readState(standbyHandle.worktreePath) and writeState(projectPath, state) but
then immediately continues, skipping the normal nextPhase() transition and
leaving the loop stuck on "eval"; after successfully reading and writing
standbyState (i.e., after writeState(projectPath, state)) call the same
nextPhase() used in the normal path (the nextPhase(...) function that advances
the state) so the state advances out of "eval" before the continue; reference
standbyHandle.worktreePath, readState, writeState, state, projectPath and
nextPhase() when making the change.
---
Nitpick comments:
In @.claude/hooks/claim-verifier.sh:
- Line 73: Replace the complex sed-based extraction of BODY from MATCH with bash
parameter expansion: strip the leading "--body" prefix from MATCH using
${MATCH#...} to remove up to the opening single-quote, then strip the trailing
single-quoted suffix using ${VAR%...} to remove from the last single-quote,
assign the result to BODY, and then normalize newlines and trim trailing
whitespace (e.g., via tr and a final sed/parameter expansion). Update the code
around the BODY and MATCH handling in claim-verifier.sh to use these expansions
instead of the long sed expression.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
Run ID: 26fbd2ea-1779-40c4-a7fa-f43bf36f52f1
⛔ Files ignored due to path filters (1)
pnpm-lock.yamlis excluded by!**/pnpm-lock.yaml
📒 Files selected for processing (16)
.beads/issues.jsonl.claude/hooks/claim-verifier.sh.github/workflows/ci.yml.gitignoreautonomous-harness/package.jsonautonomous-harness/src/cli.tsautonomous-harness/src/harness-state.tsautonomous-harness/src/index.tsautonomous-harness/src/orchestrator.tsautonomous-harness/tsconfig.jsoneslint.config.jsfile_domains.yamlpackages/cli/package.jsonpackages/cli/src/program.tspackages/core/src/__tests__/wholesome.test.tspnpm-workspace.yaml
|
@codex @coderabbitai @cursor @copilot [AI automation] Codex will implement the code updates while coderabbitai, cursor, and copilot focus on review support. Please make the following changes to this PR. Summary (Execution Flow):
PR Details:
Instructions: Tasks:
Automation Markers:
|
|
Tip For best results, initiate chat on the files or code changes. 🧩 Analysis chain🏁 Script executed: #!/bin/bash
# Fetch PR review comments
gh pr view 641 --repo jleechanorg/agent-orchestrator --json reviews,comments,reviewRequests 2>/dev/null | jq '.'
echo "---REVIEW COMMENTS---"
gh api repos/jleechanorg/agent-orchestrator/pulls/641/comments --paginate 2>/dev/null | jq '[.[] | {id, user: .user.login, body: .body, path: .path, line: .line, html_url: .html_url, created_at: .created_at}]'
echo "---ISSUE COMMENTS---"
gh api repos/jleechanorg/agent-orchestrator/issues/641/comments --paginate 2>/dev/null | jq '[.[] | {id, user: .user.login, body: .body, html_url: .html_url, created_at: .created_at}]'Length of output: 104413
📋 Consolidated Review Summary — PR
|
| Check | Status | Action Needed |
|---|---|---|
| Docstring Coverage | ❌ 39.13% (threshold: 80%) | Add JSDoc docstrings to exported functions/types in orchestrator.ts, harness-state.ts, cli.ts |
| Evidence Gate | ❓ Inconclusive | Add required sections to the PR description: Claim class, Repro gist, Terminal test output, Terminal media, UI media, Verdict |
💤 Nitpick (Low Priority — ACKNOWLEDGED)
.claude/hooks/claim-verifier.shLine 73: Consider replacingsedwith bash parameter expansion (${variable//search/replace}) per ShellCheck SC2001. Low value; defer if needed.
Summary for Codex: 7 substantive bugs to fix (items 1–7 above), 2 pre-merge check gaps (docstrings + Evidence Gate description), 1 acknowledged nitpick. All inline comment URLs are included above for the responses.json / [codex-api-automation-commit] tracking commit.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 2ff3f31230
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".
| // PRE-SPAWN standby worker for next phase (multi-worker pipelining) | ||
| const standbyHandle = await spawnStandbyWorker(phase, state, { | ||
| generatorModel, | ||
| evaluatorModel, | ||
| orchestratorModel, | ||
| skillRoot, | ||
| runtime, | ||
| projectId, | ||
| }, projectPath); |
There was a problem hiding this comment.
When pipelining is enabled, this block spawns the standby worker but never copies harness_state.json or prior sprint artifacts into standbyHandle.worktreePath. The standby prompt tells the worker to poll ./harness_state.json in its own worktree, and pollUntilActive() also reads that same worktree, so for normal worktree sessions the standby never observes the main worktree's phase transition and the harness burns the full maxIterationsPerPhase timeout before falling back to a normal spawn.
Useful? React with 👍 / 👎.
| const fullPrompt = `${config.systemPrompt}\n\nTask: ${config.taskPrompt}`; | ||
| const agentPlugin = config.model.split("/")[0]; | ||
|
|
||
| const registry = createPluginRegistry(); | ||
| await registry.loadBuiltins(config_); | ||
| const sm = await createSessionManager({ config: config_, registry }); | ||
| const session = await sm.spawn({ | ||
| projectId, | ||
| prompt: fullPrompt, | ||
| agent: agentPlugin, | ||
| runtimeOverride: config.runtime ?? "process", | ||
| skipPrBoilerplate: true, |
There was a problem hiding this comment.
The CLI exposes --generator-model, --evaluator-model, and --orchestrator-model, but this code only uses the string before / as the AO agent plugin and passes no model override to sm.spawn(). For example, --generator-model codex/gpt-5.4 selects the codex plugin but still runs whatever model is configured for the project/defaults, so the advertised model options are silently ignored.
Useful? React with 👍 / 👎.
jleechanao
changed the title
|
@coderabbitai all good? |
1 similar comment
|
@coderabbitai all good? |
|
Tip For best results, initiate chat on the files or code changes. 🧩 Analysis chain🏁 Script executed: #!/bin/bash
echo "=== PR STATUS ==="
gh pr view 641 --repo jleechanorg/agent-orchestrator --json state,mergeable,statusCheckRollup,headRefName,headRefOid 2>/dev/null | jq '.'
echo ""
echo "=== OPEN REVIEW COMMENTS ==="
gh api repos/jleechanorg/agent-orchestrator/pulls/641/comments --paginate 2>/dev/null \
| jq '[.[] | select(.resolved == false or .resolved == null) | {id, user: .user.login, body: (.body | .[0:120]), path: .path, line: .line, html_url: .html_url}]'
echo ""
echo "=== RECENT COMMITS ==="
gh api repos/jleechanorg/agent-orchestrator/pulls/641/commits --paginate 2>/dev/null \
| jq '[.[] | {sha: .sha[0:8], message: .commit.message | split("\n")[0], date: .commit.committer.date}] | .[-5:]'
echo ""
echo "=== CI CHECKS ==="
gh pr checks 641 --repo jleechanorg/agent-orchestrator 2>/dev/nullLength of output: 8375
Three things still need attention before this is merge-ready:
Once those three are done, it should be ready. 🐇 |
|
Tip For best results, initiate chat on the files or code changes. 🧩 Analysis chain🏁 Script executed: #!/bin/bash
# Get current PR status and checks
gh pr view 641 --repo jleechanorg/agent-orchestrator --json state,statusCheckRollup,headRefOid,commits 2>/dev/null | jq '{state, headRefOid, checkSummary: [.statusCheckRollup[]? | {name: .name, status: .status, conclusion: .conclusion}]}'
echo "---RECENT ISSUE COMMENTS (last 5)---"
gh api repos/jleechanorg/agent-orchestrator/issues/641/comments --paginate 2>/dev/null | jq '[ .[] | {id, user: .user.login, body: (.body | .[0:200]), created_at}] | sort_by(.created_at) | .[-5:]'Length of output: 2261 🏁 Script executed: #!/bin/bash
# Check recent commits on the PR branch
gh api repos/jleechanorg/agent-orchestrator/pulls/641/commits 2>/dev/null | jq '[.[] | {sha: .sha[0:8], message: .commit.message | split("\n")[0], date: .commit.author.date}] | .[-8:]'
echo "---KEY FILE SNIPPETS---"
# Check BODY initialization in claim-verifier.sh
echo "=== claim-verifier.sh lines 65-90 ==="
gh api repos/jleechanorg/agent-orchestrator/contents/.claude/hooks/claim-verifier.sh?ref=6ad1b1ae9daaa5771a2931eee68ff16bf63aa53b 2>/dev/null | jq -r '.content' | base64 -d | sed -n '65,90p'
echo "=== package.json exports ==="
gh api repos/jleechanorg/agent-orchestrator/contents/autonomous-harness/package.json?ref=6ad1b1ae9daaa5771a2931eee68ff16bf63aa53b 2>/dev/null | jq -r '.content' | base64 -d
echo "=== orchestrator.ts model selection lines 90-105 ==="
gh api repos/jleechanorg/agent-orchestrator/contents/autonomous-harness/src/orchestrator.ts?ref=6ad1b1ae9daaa5771a2931eee68ff16bf63aa53b 2>/dev/null | jq -r '.content' | base64 -d | sed -n '88,108p'
echo "=== orchestrator.ts local nextPhase variable (search) ==="
gh api repos/jleechanorg/agent-orchestrator/contents/autonomous-harness/src/orchestrator.ts?ref=6ad1b1ae9daaa5771a2931eee68ff16bf63aa53b 2>/dev/null | jq -r '.content' | base64 -d | grep -n "const nextPhase\|const currentPhase"
echo "=== cli.ts CommanderLike interface ==="
gh api repos/jleechanorg/agent-orchestrator/contents/autonomous-harness/src/cli.ts?ref=6ad1b1ae9daaa5771a2931eee68ff16bf63aa53b 2>/dev/null | jq -r '.content' | base64 -d | sed -n '1,30p'Length of output: 7027
✅ Confirmed Fixed (from code inspection)
|
| } else { | ||
| state = createInitialState(projectPath, projectName, totalSprints, generatorModel, evaluatorModel, orchestratorModel); | ||
| writeState(projectPath, state); | ||
| } |
There was a problem hiding this comment.
Resume ignores sprint count flag
Medium Severity
When harness_state.json already exists, the harness loads it and never applies --sprints from the CLI. A rerun with a higher sprint count still stops using totalSprints stored in the file, so users cannot extend a run via flags alone.
Reviewed by Cursor Bugbot for commit 6ad1b1a. Configure here.
|
|
||
| // If eval completed (verdict written), advance to next phase | ||
| if (phase === "eval" && state.currentSprint.verdict) { | ||
| state = nextPhase(state); |
There was a problem hiding this comment.
Failed eval still advances
Medium Severity
Any truthy verdict, including "fail", triggers nextPhase, which appends the sprint to completedSprints and starts the next sprint or finishes the run. There is no branch to stop or retry after a failed evaluation.
Additional Locations (1)
Reviewed by Cursor Bugbot for commit 6ad1b1a. Configure here.
| throw new Error( | ||
| `[autonomous-harness] Phase ${phase} did not advance after worker completion — treating as run failure.`, | ||
| ); | ||
| } |
There was a problem hiding this comment.
No cleanup on harness failure
Medium Severity
When phase advancement validation fails, the harness throws without terminating the active worker or pre-spawned standby session created in the same iteration. Those AO sessions keep running after the CLI process exits.
Reviewed by Cursor Bugbot for commit 6ad1b1a. Configure here.
There was a problem hiding this comment.
Actionable comments posted: 3
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@autonomous-harness/src/cli.ts`:
- Line 19: Change the action signature to remove both `any` types and use a
generic for the options type so callers can call `cmd.opts<T>()`; e.g. replace
the current declaration of `action(handler: (flags?: any, cmd?: any) =>
unknown): CommanderLike;` with a generic form such as `action<T = Record<string,
unknown>>(handler: (flags?: T, cmd?: { opts: () => T }) => unknown):
CommanderLike;` so the handler signature reflects the actual usage of
`cmd.opts<T>()` and eliminates `any`.
In `@packages/core/src/__tests__/harness.test.ts`:
- Line 2: Replace the relative import to the harness state with the workspace
package name: import createInitialState and nextPhase from
"`@jleechanorg/ao-autonomous-harness`" instead of
"../../../../autonomous-harness/src/harness-state.ts"; also ensure
packages/core/package.json lists "`@jleechanorg/ao-autonomous-harness`" as a
devDependency so the package resolution uses the built package rather than a src
path.
- Around line 4-21: Extend the existing test in harness.test.ts to assert that
calling nextPhase on the state in "eval" advances to "done" and then verify the
behavior of nextPhase when the current sprint is already "done" (either remains
"done", throws, or cycles based on intended design); locate and update the test
using createInitialState and nextPhase to add an additional expect for
state6.currentSprint.phase === "done" and a final assertion
describing/validating the expected outcome of invoking nextPhase on the done
state.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
Run ID: 5c83482f-f8c9-4c79-bc3c-a91dfd64c1e9
📒 Files selected for processing (6)
.claude/hooks/claim-verifier.shautonomous-harness/package.jsonautonomous-harness/src/cli.tsautonomous-harness/src/orchestrator.tspackages/core/src/__tests__/harness.test.tspackages/core/src/__tests__/wholesome.test.ts
🚧 Files skipped from review as they are similar to previous changes (3)
- autonomous-harness/package.json
- .claude/hooks/claim-verifier.sh
- autonomous-harness/src/orchestrator.ts
|
@coderabbitai all good? |
|
Tip For best results, initiate chat on the files or code changes. 🧩 Analysis chain🏁 Script executed: #!/bin/bash
echo "=== PR STATUS ==="
gh pr view 641 --repo jleechanorg/agent-orchestrator --json state,mergeable,statusCheckRollup,headRefOid 2>/dev/null | jq '{state, mergeable, headRefOid, checks: [.statusCheckRollup[]? | {name: .name, status: .status, conclusion: .conclusion}]}'
echo ""
echo "=== RECENT COMMITS ==="
gh api repos/jleechanorg/agent-orchestrator/pulls/641/commits 2>/dev/null | jq '[.[] | {sha: .sha[0:8], message: (.commit.message | split("\n")[0]), date: .commit.author.date}] | .[-6:]'
echo ""
echo "=== OPEN REVIEW THREADS ==="
gh api repos/jleechanorg/agent-orchestrator/pulls/641/comments --paginate 2>/dev/null | jq '[.[] | select(.resolved == false or .resolved == null)] | length'
echo ""
echo "=== spawnStandbyWorker nextPhase shadow check ==="
gh api "repos/jleechanorg/agent-orchestrator/contents/autonomous-harness/src/orchestrator.ts" 2>/dev/null | jq -r '.content' | base64 -d | grep -n "const nextPhase\|const nextPhaseName\|const targetPhase\|const currentPhase" | head -20
echo ""
echo "=== standby seeding check (harness_state copy into standby) ==="
gh api "repos/jleechanorg/agent-orchestrator/contents/autonomous-harness/src/orchestrator.ts" 2>/dev/null | jq -r '.content' | base64 -d | grep -n "standby\|copyFileSync\|cp -r\|harness_state" | head -30
echo ""
echo "=== standby nextPhase guard check (lines 700-750 area) ==="
gh api "repos/jleechanorg/agent-orchestrator/contents/autonomous-harness/src/orchestrator.ts" 2>/dev/null | jq -r '.content' | base64 -d | sed -n '700,760p'Length of output: 8351
✅ Confirmed Fixed
❌ Still Outstanding1. Merge conflicts — CONFLICTING 2. // Line 87 — still shadows the imported nextPhase function
const nextPhase: Phase = PHASE_ORDER[nextPhaseIdx]!;Rename to 3. 27 open review threads on GitHub 4. Cursor Bugbot — still Summary
Once Codex clears the shadowing rename, resolves conflicts, and runs |
…res, and path validation
jleechan2015
force-pushed
the
feat/autonomous-harness-impl
branch
from
21bbf4a to
0e4b19b
Compare
|
SKEPTIC_GATE_TRIGGER |
jleechan2015
changed the title
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 2 potential issues.
There are 5 total unresolved issues (including 3 from previous reviews).
Bugbot Autofix is ON. A cloud agent has been kicked off to fix the reported issues.
Reviewed by Cursor Bugbot for commit 0e4b19b. Configure here.
|
|
||
| ## References | ||
| - ~/roadmap/ts-harness-pr-513-next-steps-2026-05-03.md | ||
| - PR: https://github.com/jleechanorg/agent-orchestrator/pull/513", "status": "open", "priority": 1, "issue_type": "task", "created_at": "2026-05-03T02:30:30.133668+00:00", "created_by": "jleechan", "updated_at": "2026-05-03T02:30:30.133807+00:00", "closed_at": null, "close_reason": null, "source_repo": ".", "compaction_level": 0, "original_size": 0, "labels": ["autonomous-harness", "typescript"]} |
There was a problem hiding this comment.
Invalid multiline beads JSONL record
Medium Severity
The appended bd-ts01 issue spans multiple physical lines with raw line breaks inside the description string instead of escaped \n, so it is not valid JSON on one JSONL line. Line-by-line beads importers fail on this record, and the same bd-ts01 id already exists earlier in the file as a valid single-line entry.
Reviewed by Cursor Bugbot for commit 0e4b19b. Configure here.
| "zod": "^3.24.0" | ||
| }, | ||
| "devDependencies": { | ||
| "@jleechanorg/ao-autonomous-harness": "workspace:*", |
There was a problem hiding this comment.
Core tests need harness build
Medium Severity
packages/core now depends on @jleechanorg/ao-autonomous-harness and imports it from harness.test.ts, but its test script has no step to build that workspace package first. Running pnpm --filter @jleechanorg/ao-core test without a prior harness build fails resolving dist/index.js.
Reviewed by Cursor Bugbot for commit 0e4b19b. Configure here.
|
SKEPTIC_GATE_TRIGGER |
Green Gate — Deterministic 6-Green Check
Result: FAIL — gates 1 failed VERDICT: FAIL |
2 participants
Summary
file_domains.yamlwith domain lock registry forskeptic-core,skeptic-cli, andci-workflows(advisory) domains.codex/,.gemini/,.opencode.json)Part of the merge_train integration across agent repos. Enables AI agents to check and reserve file domains before editing, preventing concurrent PR conflicts on core skeptic files.
Governing doc: N/A — cross-repo tooling install
Bead: N/A
Test plan
domain_lock audit --registry file_domains.yamlparses cleanlydomain_lock check --files packages/core/src/skeptic-reviewer.tsresolves toskeptic-core🤖 Generated with Claude Code
Note
Medium Risk
Domain registry changes affect which PRs collide on skeptic-critical paths; orchestrator path mismatch is a new startup failure mode for misconfigured workspaces.
Overview
Extends
file_domains.yamlwithskeptic-coreandskeptic-cliglobs so merge_train / area-lock can treat core skeptic and CLI paths as collision domains at spawn time..gitignorenow ignores merge_train install artifacts (whole.gemini/,.codex/,.opencode.json) instead of only a single Gemini script path.Autonomous harness package entrypoints point the default export at
dist/index.js(library API) while keeping./cliseparate; the orchestrator fails fast when resolved AO projectpathdoes not match--project-path.@jleechanorg/ao-coreadds a workspace devDependency on the harness and a Vitest check that sprint phases advance research → … → done.A beads tracking row was added for the TS harness push/cleanup work (metadata only).
Reviewed by Cursor Bugbot for commit 0e4b19b. Bugbot is set up for automated code reviews on this repo. Configure here.
Summary by CodeRabbit
New Features
ao autonomous-harnessCLI to run sprint-based autonomous workflows (research → plan → annotation → implementation → eval → done) with configurable options and multi-worker pipelining.Tests
Chores