Delete all archived packages from legacy-packages/.
These were preserved for reference in the old repo but are
not needed in the new united-workforce repo.
Ref: shazhou/united-workforce#7
Problems on macOS:
- `-v $HOME:$HOME` let container's bun install overwrite host bun
binary (Linux ARM64 replaced macOS ARM64)
- Container couldn't reach host LLM endpoints (localhost != host)
- Hardcoded `~/repos/workflow` path didn't exist on all machines
Fixes:
- Mount source read-only (`-v $(pwd):/workspace:ro`) + copy inside
- Use container-local HOME (/root) to isolate bun/npm installs
- Add `--add-host=host.docker.internal:host-gateway` for Linux compat
- `docker cp` host config + sed localhost→host.docker.internal
- Use `debate.yaml` instead of `solve-issue.yaml` (no $SUSPEND dep)
- Fix cancel test: `--status cancelled` not `--status completed`
Verified: full 6-step walkthrough passes on macOS, host bun intact.
The $SUSPEND edge for insufficient_info uses {{{reason}}} template
variable, but the frontmatter schema was missing the reason field.
This caused workflow validation to reject the workflow on thread start.
Fixed in all 3 copies: .workflows/, examples/, workflows/
- .workflows/solve-issue.yaml
- examples/solve-issue.yaml
- workflows/solve-issue.yaml
All planner insufficient_info routes now use $SUSPEND instead of $END.
Closes#592
- thread list: suspended threads show [suspended] marker via statusDisplay
- thread show: displays suspendedRole, suspendMessage, and resume hint
- New ThreadShowOutput type with hint field
- Tests: 3 cases for display formatting
Closes#591
- Add GraphPseudoRole type ($END | $SUSPEND) to workflow-protocol
- Add 'suspended' to ThreadStatus
- evaluate() returns EvaluateSuspendResult for $SUSPEND targets
- Thread show/list derive suspended status from moderator evaluation
- validate-semantic treats $SUSPEND like $END (valid target, no outgoing edges)
- Tests: routing to $SUSPEND, mustache rendering, thread status display
Closes#588
- Add !include custom YAML tag for referencing external files (Fixes#582)
- .md/.txt files included as strings
- .json files parsed as JSON objects
- .yaml/.yml files parsed as YAML objects
- Paths resolved relative to the workflow YAML file
- Support foo/index.yaml as alternative to foo.yaml (Fixes#583)
- Updated discoverProjectWorkflows(), findWorkflowInDir()
- Updated workflowNameFromPath() for index.yaml detection
- Flat files take priority over folder layout
- Added tests for both features
Fixes hallucination issues observed in thread 06F7FSTXQGY3D5CY5YPQFK2Y3W:
1. Developer self-verification (critical): Added step 12 requiring
mandatory verification of branch, file existence, and git status
before reporting done status. Prevents hallucinated completions
without actual tool execution.
2. Reviewer hard-check enforcement (critical): Added critical warning
and step 0 requiring cd/pwd verification before review. Prevents
false rejections based on assumptions without actual path checks.
3. Test debugging escalation (medium): Added structured debugging
guidance with escalation path after 3 test cycles. Prevents
infinite retry loops by providing strategy and fail-fast guidance.
Also added 3 test cases to verify the new procedure steps exist.
Based on change plan 9EVZPDTS16PMG analyzing execution anomalies
that resulted in 58% waste (13 of 23 minutes).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Fixes hallucination issues observed in thread 06F7FSTXQGY3D5CY5YPQFK2Y3W:
1. Developer self-verification (critical): Added step 12 requiring
mandatory verification of branch, file existence, and git status
before reporting done status. Prevents hallucinated completions
without actual tool execution.
2. Reviewer hard-check enforcement (critical): Added critical warning
and step 0 requiring cd/pwd verification before review. Prevents
false rejections based on assumptions without actual path checks.
3. Test debugging escalation (medium): Added structured debugging
guidance with escalation path after 3 test cycles. Prevents
infinite retry loops by providing strategy and fail-fast guidance.
Also added 3 test cases to verify the new procedure steps exist.
Based on change plan 9EVZPDTS16PMG analyzing execution anomalies
that resulted in 58% waste (13 of 23 minutes).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Remove --repo flag from tea pr create (fails in git worktrees)
- Add guard to skip staging when developer already committed changes
Co-Authored-By: Claude Opus 4 <noreply@anthropic.com>
Handle the case where Claude Code exits without producing a result line
(timeout, OOM, signal kill). Previously returned null and threw an error;
now returns incomplete result with best-effort output extraction.
Changes:
- Add "incomplete" as new ClaudeCodeResultSubtype value
- Extract output from last assistant turn when no result line exists
- Enhanced error messages distinguish incomplete vs unparseable output
- Store incomplete results in CAS with appropriate metadata
- Add 10 comprehensive test cases for incomplete result handling
Fixes#574
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This resolves issue #573 by moving uwf's CAS directory from
~/.uncaged/workflow/cas/ to the shared ~/.uncaged/json-cas/ location.
Changes:
- Added getGlobalCasDir() function with UNCAGED_CAS_DIR support
- Updated createUwfStore() to use global CAS directory
- Added comprehensive test coverage (11 new tests)
- Updated all existing tests for environment isolation
- Updated documentation (CLAUDE.md, README.md)
Benefits:
- Cross-tool visibility: json-cas CLI can read uwf-created nodes
- Schema sharing: both tools access same schema registry
- Future-proofing: enables json-cas render/verbose for uwf data
Fixes#573
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This resolves issue #573 by moving uwf's CAS directory from
~/.uncaged/workflow/cas/ to the shared ~/.uncaged/json-cas/ location.
Changes:
- Added getGlobalCasDir() function that respects UNCAGED_CAS_DIR env var
- Updated createUwfStore() to use the global CAS directory
- Updated all tests to set UNCAGED_CAS_DIR for test isolation
- Added comprehensive test suite for global CAS functionality
- Updated documentation (CLAUDE.md, README.md) to reflect new architecture
Benefits:
- Cross-tool visibility: json-cas CLI can now read uwf-created nodes
- Schema sharing: both tools access the same schema registry
- Future-proofing: enables json-cas render/verbose features for uwf data
Workflow metadata (threads.yaml, registry.yaml, history.jsonl) remains in
~/.uncaged/workflow/ as intended.
All tests pass. No breaking changes to existing functionality.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add GROUND RULES section to all role procedures: require real command output, no fabrication
- Add 'skipped' status for roles where everything is already configured
- Add skipped routing in graph so workflow continues normally
- Add build artifact detection in committer: scan for .d.ts/.js.map/.js before commit
- Add verification enforcement notes to all roles
Fixes hallucination issue where agents reported completing work without actually writing files.
Store the fully assembled prompt sent to each agent in CAS as a text
node, referenced from StepNodePayload.assembledPrompt. This enables
exact reproduction of what the agent received for debugging hallucinations.
Changes:
- workflow-protocol: StepRecord + STEP_NODE_SCHEMA add assembledPrompt field
- workflow-util-agent: AgentRunResult includes assembledPrompt, run.ts stores it
- workflow-util-agent: schemas register TEXT_SCHEMA for prompt storage
- workflow-agent-claude-code: return assembled prompt from buildClaudeCodePrompt
- workflow-agent-hermes: return assembled prompt from buildHermesPrompt
- workflow-agent-builtin: return empty prompt (no prompt assembly)
- cli-workflow: step read --prompt renders the stored prompt
- All test fixtures updated for new field
Legacy steps without assembledPrompt show 'Prompt not recorded' message.
小橘 🍊
- Add git ls-remote verification after push
- Switch from tea pr create to Gitea API (tea fails in worktrees)
- Add PR creation verification (check response JSON has number field)
- Explicitly mark steps 4 and 6 as verification gates
小橘 🍊
Root cause: committer role had to parse owner/repo from git remote URL,
which failed in worktrees with token-embedded URLs. Agent hallucinated
a fake PR URL instead of reporting the error.
Fix:
- planner extracts repoRemote from git remote, stores in frontmatter
- repoRemote flows through all roles via graph prompts
- committer uses repoRemote directly for tea/API calls
- Added Gitea API fallback when tea CLI fails
小橘 <xiaoju@shazhou.work>
xiaomo
xingyue
xiaoju
xiaonuo