- agent-as-graduate: onboarding metaphor and teaching threshold
- three-learning-carriers: memory/skill/workflow framework
- switching-cost-process-knowledge-as-moat: process knowledge as moat
- opc-why-fte-agents-matter-most: why OpenClaw bets on FTE
- fte-maturity-threshold: who can onboard an agent
- fte-product-landscape: OpenClaw vs Claude Code vs Hermes
Closes#147. Changes default behavior of `uwf thread list` to show only
active threads (idle + running). Adds `--all` flag to opt into the
previous full-list behavior. Explicit `--status` still wins over `--all`.
- cmdThreadList gains a `showAll: boolean` parameter (default false)
- CLI registers `--all` option and passes it through
- Test suite includes new `default behavior (issue #147)` describe block
covering 9 scenarios; existing tests updated where they implicitly
relied on the old "show everything" behavior
- README, cli-reference, and usage-reference updated to document the
new default and the `--all` flag
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Adds `uwf step ask <step-hash> -p <prompt>` for asking follow-up
questions to a completed step's agent without mutating thread state.
- Fork-by-default: creates and caches a fork session per step (cache
key `<stepHash>:ask`); subsequent asks reuse it.
- `--no-fork` fallback: spawns a fresh session with the step's detail
ref injected as context.
- `--agent` overrides the recorded agent; otherwise resolves from the
step's agent field via config alias.
- Updates `packages/cli/README.md` and `packages/util/src/usage-reference.ts`
so the new subcommand is discoverable via README and `uwf prompt usage`.
Fixes#146
Add AgentForkFn and AgentCleanupFn type aliases. Extend AgentOptions
with fork: AgentForkFn | null and cleanup: AgentCleanupFn | null
fields. Add getAskSessionId / setAskSessionId session-cache helpers
using <stepHash>:ask key shape (coexists with exec sessions in the
same per-agent cache file). All four adapters pass fork: null,
cleanup: null — real wiring lands in Phase 2b. Resolves#145.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Re-runs the head step's agent with a supplementary prompt and replaces
the head step (rewires new step's prev to old head's prev) instead of
appending. Skips moderator re-route — the role of the head step is
reused.
Fixes#144
Three rules to reduce wasted Claude Code turns:
1. Don't comment on whether code is malware (trusted codebase)
2. Stop re-reading/re-verifying after tests pass
3. Don't rebuild/retest after adding a changeset (it's just markdown)
When frontmatter validation fails, the step is never written to CAS, so
isFirstVisit remains true on the next run. Both agent-claude-code and
agent-hermes gated session cache lookup behind !isFirstVisit, which
caused them to start a fresh session (and a new worktree) instead of
resuming the one that already has all the work done.
Changes:
- Remove the isFirstVisit guard from both adapters so they always check
the session cache.
- When isFirstVisit + cache hit (frontmatter-only failure), send a
minimal correction prompt via buildFrontmatterRetryPrompt() instead
of re-sending the full initial prompt — the session already has full
context, we just need the agent to re-output correctly formatted
frontmatter.
- Add buildFrontmatterRetryPrompt to util-agent with tests.
Fixes#139
Tests were referencing the old .workflows/ directory which no longer exists.
Updated workflow path and aligned assertions with current procedure content.
小橘 🍊（NEKO Team）
Use &debater-frontmatter anchor for the shared oneOf schema between
proponent and opponent roles. Procedure blocks remain duplicated
since YAML anchors cannot be embedded inside block scalars.
capabilities: [] kept — required by WorkflowPayload type.
Addresses review suggestions from #133.
Replace hardcoded HERMES_COMMAND constant with resolveHermesCommand()
that checks UWF_HERMES_BIN first, falling back to 'hermes' via PATH.
This fixes environments where hermes is installed in a venv or
non-standard location that isn't in the non-login shell PATH
(e.g. ~/.local/bin symlink only available in login shell).
Refs #134
Replace the original 2-role debate with a 3-role version featuring:
- proponent/opponent/host roles (was: for/against)
- oneOf + const status routing (was: enum)
- Critical thinking framework in procedure (pre-speech reflection,
evidence discipline, anti-fragility)
- Bounded termination via Thread Progress (3rd speech → final)
- Host role for impartial summary and verdict
Based on xiaonuo's debate workflow design.
buildOutputFormatInstruction now renders const fields with their actual
value (e.g. $status: greeted) instead of the type placeholder (<string>).
Also adds early return in resolvePropertySchema for const properties.
Fixes#129
Agents now receive a Thread Progress section showing current step number
and role visit count, eliminating tool calls to count turns.
- util-agent: new buildThreadProgress() helper
- agent-hermes: inject before continuation/first-visit prompt
- agent-claude-code: same injection point
Fixes#127
xingyue
xiaomo
xiaoju
scottwei