Bundle top-level code runs during `workflow add` (descriptor extraction),
but agent config env vars (e.g. WORKFLOW_HERMES_COMMAND) are only available
at `workflow run` time. Deferring validation prevents premature throws.
BREAKING: HermesAgentConfig.command and CursorAgentConfig.command are now
required string fields (absolute path to CLI binary). Validation rejects
non-absolute paths at construction time.
- Eliminates PATH resolution ambiguity in spawned worker processes
- spawnCli: explicit env: process.env for clarity
- bundle-entry: WORKFLOW_CURSOR_COMMAND is now required
- Updated tests for both agents
- scripts/publish-all.sh: bun pm pack (resolves workspace:*) + npm publish
- All 14 public @uncaged/* packages published to git.shazhou.work
- CLAUDE.md: document Gitea registry, bunfig.toml scoped registry, publish workflow
- bun link docs demoted to alternative for un-published local changes
- Add link/link:consume/link:unlink scripts to root package.json
- Document cross-repo bun link workflow in CLAUDE.md
- Fix hardcoded @uncaged/workflow-runtime "^0.1.0" → "*" in
init workspace and init template scaffolds (actual version is 0.3.x)
- Workflow cards click to expand/collapse
- Lazy-load descriptor on first expand
- Static WorkflowGraph (all nodes default state, no highlighting)
- Show description, version count, hash
- Fix WorkflowSummary type to match actual API response
- bindingInitializerIsCallable: accept Identifier (e.g. var run = wf)
- import allowlist: startsWith('@uncaged/workflow') instead of exact match list
小橘 🍊（NEKO Team）
Design doc for parent-child workflow Merkle linking:
- StartNodePayload.parentState: child → parent head state at spawn time
- StateNodePayload.childThread: parent → child final state hash
- Both also in refs[] for GC reachability
- 4-phase implementation plan
小橘 <xiaoju@shazhou.work>
1. Replace manual OpenAI response parsing with createThreadReactor —
workspace extraction now uses the same ReAct loop as extract/summarizer,
with a zod schema and structured tool call
2. Remove non-null assertion on llmProvider, replaced with explicit guard
3. Add structured logging (LogFn) to extraction — failures, non-absolute
paths, and successful extractions all logged with tag V3KM8QWP
4. export const run = wf is correct: createWorkflow returns WorkflowFn
directly, old bundle-entry had stale .run access + unused 3rd arg
小橘 <xiaoju@shazhou.work>
Replace 50ms setTimeout with waitUntilPredicate polling for first role
completion before issuing kill. Same pattern used by pause/resume test.
小橘 <xiaoju@shazhou.work>
- Sort threads by startedAt descending (newest first)
- completed: green (dimmed) — distinct from running (green, full opacity)
- active: accent color (blue) — was same muted gray as completed
- failed: red — unchanged
- running: green — unchanged
Fixes#191
v0.3.0 was published with workspace:* deps (npm doesn't resolve them).
Re-published as v0.3.1 with concrete version refs.
Local monorepo deps restored to workspace:*.
小橘 <xiaoju@shazhou.work>
Deduplicate CAS_GET_TOOL_DEFINITION, isRecord, toolHandler, and
structuredToolFromSchema between summarizer.ts and extract-fn.ts.
Both now use createCasReactor(provider, cas, opts) and only provide
their own systemPrompt.
小橘 <xiaoju@shazhou.work>
Instead of hardcoding 'completed: moderator returned END', the engine now
calls a summarizer (ReAct loop with cas_get tool) to produce a meaningful
summary of the workflow outcome before writing the __end__ node.
- New summarizer.ts following supervisor.ts pattern
- Uses extract scene LLM provider (falls back to raw completion.summary on failure)
- Tracks step contentHashes for summarizer context
- Schema: z.object({ summary: z.string() })
Refs #187, #188
小橘 <xiaoju@shazhou.work>
hasRoundsRemaining is now always true — supervisor controls when to
stop, not a round counter. Tests updated to expect coder retry
instead of END on exhaustion.
Refs #185
xingyue
xiaomo
xiaoju