- Wrap skill copying in copy_skills() helper function
- Call copy_skills() before early exit for lightweight states (Triage/Closure/Request Changes)
- Clone openclaw repo BEFORE copying skills (fixes empty-dir clone failure)
- Call copy_skills() AFTER clone + checkout for normal states
- Replace 'git pull --rebase origin HEAD' with 'git rebase origin/main' in before_run
- Add regression test to verify hook ordering in core_test.exs
Local clone for speed, then reset origin to github.com/openclaw/openclaw.
Add git fetch origin to before_run hook so repeat Prepare runs
get fresh origin/main refs instead of stale local-clone state.
Side-lane state that rebases PR branch onto main, resolves mechanical
conflicts, force pushes, and drops back to Todo for human decision.
No review, no gates, no tests -- just freshens the branch.
Linear state: de50ceb9-a0ef-4f13-849f-bf31a65392ee
- Created Test state (591e5db0) and Pre-merge state (3f6e88cf) in Linear
- Test is an optional lane: Prepare Complete -> Test -> Pre-merge -> Merge
- Josh can also skip Test and go Prepare Complete -> Merge directly
- Test phase runs full suite, distinguishes pre-existing vs PR-introduced failures
- Pre-merge is a human gate with notification
- Max 1 concurrent test agent (resource constraint, same as prepare)
- WORKFLOW.md: Prepare phase now generates executable test scripts in
.local/test-kit/ for PRs with API-observable behavioral changes
- common.sh template with gw_call() and assert_jq() helpers
- Numbered test scripts per scenario, independently runnable
- Linear comment includes worktree path to test-kit
- priv/scripts/test-gateway.sh: standalone tool to run a PR build
on alternate port (18790) without touching the production gateway
Supports start/stop/status/call subcommands
After fixes and gates pass, the prepare agent now generates .local/test-plan.md
with concrete verification steps executable on a live OpenClaw instance.
Test plans include:
- Risk tier classification (GREEN/YELLOW/RED)
- Specific CLI commands and expected outcomes
- Rollback instructions
- Pre-test checklist for branch checkout/build/restart cycle
- Untestable changes section for transparency
Review and Prepare agents now query issue comments before starting
work. Maintainer can leave fix directions, focus areas, or constraints
on the Linear issue and agents will incorporate them.
- Move skill copy before early-exit in after_create hook (fixes pr-cluster missing for Triage agents)
- Add Triage and Request Changes as active states
- Move enrichment template from Backlog to Triage
- Add prior GitHub CHANGES_REQUESTED review check to Triage template
- Require structured duplicate assessment comments on both duplicate paths
- Add Request Changes template (posts gh pr review, moves to Backlog)
- Add missing state IDs to states map (review, review_complete, prepare, prepare_complete)
- Default mix caclawphony.review to Triage state (--direct skips to Review)
- gitignore workspaces/
Closure agents were refusing to post GitHub closing comments because
the 'Never comment on the PR' rule appeared in a global Rules section.
Now conditionally excluded for Closure phase, which needs gh pr comment
and gh pr close to do its job.
Two bugs caused MAR-55 duplicates to get stuck:
1. Enrichment created Duplicate issues but relations weren't verified
2. Merge phase didn't move duplicates to Closure state
Fixes:
- Add relation verification step after issueRelationCreate in enrichment
- Add auto-transition of Duplicate issues to Closure in merge phase
- Use {:map, :string, :any} and {:map, :string, :string} types for labels/states
- Return empty map instead of :omit when config sections are absent
- Fix test using self() as agent pid (causes EXIT shutdown)
- Update in-repo WORKFLOW.md integration tests to match current prompt
- Fix list ordering assertions for gate states (use Enum.sort)
- Add notifications section to WORKFLOW.yaml with telegram credentials and template
- Add gates section with state_id, assignee, and notify for each gate
- Add labels section with recommendation and subsystem label UUIDs
- Add states section for state name to ID mappings
- Add retry_base_ms and continuation_delay_ms to agent config
- Config module parses all new sections with env var resolution
- Notifier reads from Config instead of env vars directly
- Orchestrator uses Config for gate states and retry delays
- PromptBuilder exposes labels/gates/states as Solid template variables
- Tests cover all new config paths
State transition to a non-active state (e.g. Todo, Review Complete)
causes the orchestrator to kill the agent. If the comment mutation
comes after the state transition, it never executes.
Reordered all four phases to: comment first, state transition last.
Linear labels created:
- Recommendation: review (green), wait (yellow), skip (gray)
- Subsystem: gateway, channels, browser, agents, config, cli,
runtime, auth, providers, docs
Enrichment agent now sets all metadata in a single issueUpdate:
- Title with [REVIEW]/[WAIT]/[SKIP] prefix
- Priority (0-4 based on recommendation + scope)
- Estimate (Fibonacci complexity 1/2/3/5/8)
- Labels (one recommendation + matching subsystem labels)
- Assignee (Josh at human gates, for review queue filtering)
Review/Prepare phases also assign to maintainer at gate transitions.
Triage task gains --priority flag for manual override on intake.
after_create hook now copies skill files from maintainers repo into
the workspace with cp -RL (resolving symlinks). Skills stay in one
place, workspaces get fresh copies, prompt just says 'read the skill'.
Codex sandbox can't read symlinked skill files from host filesystem.
Inline the full step-by-step instructions directly in the prompt template
so the agent has everything it needs without file reads.
Summary:
- remove workspaces directly from orchestrator terminal-state cleanup
- build cleanup path as Config.workspace_root() <> "/" <> issue identifier
- log workspace cleanup actions when terminal issues are detected
Rationale:
- enforce immediate workspace cleanup when Linear issues reach terminal states
- align cleanup path and deletion behavior with requested implementation
Tests:
- cd elixir && mix compile
- cd elixir && mix test (first run had one flaky timing failure; rerun passed)
Refs: caclawphony-f2f
Regeneration-Prompt: |
Implement workspace cleanup in the orchestrator when an issue reaches a
terminal state (Done, Canceled, Duplicate). Keep cleanup limited to terminal
transitions only, not active or gate states. Update the cleanup helper to
construct the workspace path from Config.workspace_root() and issue
identifier, remove it with File.rm_rf/1, and emit a log entry describing the
cleanup. Verify by compiling and running tests, then close pebbles issue
caclawphony-f2f and commit the change.
Co-authored-by: Codex <codex@openai.com>
Summary:
- Add `mix caclawphony.review` task for batch PR intake
- Fetch PR title/url via `gh pr view` per PR number
- Create Linear issues in MAR team Review state + Caclawphony project
- Print created Linear identifier and URL for each PR
Rationale:
- Provide a direct CLI for converting PR numbers into Linear review work
- Keep Linear GraphQL interactions aligned with existing client patterns
Tests:
- cd elixir && mix compile
- cd elixir && LINEAR_API_KEY=test mix caclawphony.review --help
Issue:
- caclawphony-432
Regeneration-Prompt: |
Implement a new Mix task `mix caclawphony.review` that accepts one or
more PR numbers and creates one Linear issue per PR for review intake.
Preserve existing project behavior by reusing the repository's
`SymphonyElixir.Linear.Client.graphql/3` helper instead of inventing a
new HTTP client path.
For each PR number, call GitHub CLI to fetch title exactly via
`gh pr view <num> --json title -q .title` and gather the PR URL for the
issue description. Build Linear issue titles as `PR #<num>: <title>`,
include PR metadata in the description, and set `stateId` to Review and
`projectId` to Caclawphony using the provided IDs.
Determine the team ID dynamically by querying Linear for team key `MAR`
before creating issues. Require `LINEAR_API_KEY` through existing config
resolution, fail loudly on GraphQL errors or command failures, and print
the created issue identifier for each PR.
Co-authored-by: Codex <codex@openai.com>
prepare-pr and merge-pr need gh CLI, pnpm install, and network access
that workspace-write sandbox blocks. Using full access for now.
Closes caclawphony-5fb
#### Context
Replace the custom Elixir observability TCP server with Phoenix while
keeping the shipped escript and existing operational API calls working
the same way they did before.
#### TL;DR
*Move the Elixir observability dashboard and API onto Phoenix without
breaking the escript or curl workflows.*
#### Summary
- replace the custom HTTP server with a Phoenix endpoint, router,
controller, and shared presenter
- add a LiveView operations dashboard with PubSub-driven updates and
embed its CSS and JS assets in code
- preserve runtime compatibility by accepting form-style POSTs and
extending the endpoint coverage around those paths
#### Alternatives
- keep extending the hand-rolled TCP server, but that keeps custom
parsing and routing complexity in the app
- serve dashboard assets from priv directories, but the shipped escript
cannot rely on those files existing on disk
#### Test Plan
- [x] `make -C elixir all`
- [x] `mix -C elixir test test/symphony_elixir/extensions_test.exs`
- [x] `./bin/symphony
--i-understand-that-this-will-be-running-without-the-usual-guardrails
--port 40123`
- [x] `curl -si http://127.0.0.1:40123/dashboard.css`
- [x] `curl -si http://127.0.0.1:40123/vendor/phoenix/phoenix.js`
- [x] `curl -si -X POST -d '' http://127.0.0.1:40123/api/v1/refresh`
- [x] `curl -si -X POST -d '' http://127.0.0.1:40123/api/v1/state`
---------
Co-authored-by: Codex <codex@openai.com>
