- Add checkStaleAgents() to detect agents reporting "running" but not updating
- Add markAgentDead() to update agent bead state to "dead"
- Integrate stale agent check into heartbeat cycle
- DeadAgentTimeout set to 15 minutes
This is a safety mechanism for agents that crash without updating their state.
The daemon now marks them as dead so they can be restarted.
Also fixes duplicate AgentFields declaration - now uses beads.go version with
ParseAgentFieldsFromDescription alias in fields.go.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add AgentFields and ParseAgentFieldsFromDescription to internal/beads/fields.go
- Update daemon/lifecycle.go to use shared parsing
- Update cmd/molecule_status.go to use shared parsing
- Remove duplicate parsing code and unused isAgentRunningByBead function
This consolidates agent bead field parsing in one place, following the pattern
established for AttachmentFields and MRFields.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Create ephemeral agent beads for ZFC-compliant polecat state tracking.
- Add AgentFields struct and helpers to beads package
- CreateAgentBead: Creates agent bead with role_type/rig/agent_state
- UpdateAgentState: Updates agent_state and hook_bead fields
- DeleteAgentBead: Hard-deletes ephemeral agent bead
- GetAgentBead: Retrieves and parses agent bead
- Integrate lifecycle in polecat manager:
- Add(): Creates gt-polecat-<rig>-<name> bead with state=spawning
- Recreate(): Deletes old bead, creates fresh with state=spawning
- RemoveWithOptions(): Deletes agent bead on nuke
This enables Witness to read polecat state from beads instead of
tmux scraping. State updates (spawning→working→done) are done by
polecats via bd agent state (separate beads CLI enhancement).
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add getAgentBeadState() and getAgentBeadInfo() to read agent state from beads
- Add identityToAgentBeadID() to map daemon identities to agent bead IDs
- Update ensureDeaconRunning() to check agent bead state first (ZFC compliant)
- Add agent bead state logging in executeLifecycleAction()
This is the first step toward ZFC-compliant state detection. Dependent tasks:
- gt-psuw7: Remove PID/tmux state inference
- gt-2hzl4: Add timeout fallback for dead agents
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This change migrates gt mol status from querying by pinned status + assignee
to reading the hook_bead field from agent beads.
Changes:
- Add AgentBeadFields struct to parse agent bead description fields
- Add buildAgentBeadID function to convert identity to agent bead ID
- Update runMoleculeStatus to:
1. First try to find the agent bead by ID
2. Read hook_bead from agent bead description
3. If hook_bead is set, fetch and display that bead
4. Fall back to legacy pinned-query approach if no agent bead or hook
The implementation is backwards compatible - agents without agent beads
(like polecats before gt-rxa7v is completed) still work via the fallback.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Added ResolveBeadsDir() helper that follows .beads/redirect files,
enabling crew workers and polecats to properly access shared beads.
Updated callers:
- mailbox.go: NewMailboxFromAddress follows redirect
- catalog.go: LoadCatalog follows redirect at all levels
- doctor checks: beads_check, patrol_check, wisp_check follow redirect
Also added comprehensive tests for the redirect resolution logic.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Adds a new fixable doctor check that verifies beads routing configuration:
- Checks if routes.jsonl exists
- Verifies all rigs have routing entries (by path, not just prefix)
- Validates that routes point to valid locations with .beads directories
- Can auto-fix by adding missing routes with --fix
The check is smart about prefix mismatches: if a rig already has a route
by path (e.g., gastown/mayor/rig), it won't report it as missing even if
the prefix in rigs.json differs from what's in routes.jsonl.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When mayor/deacon checks their hook from ~/gt, gt mol status now scans all
registered rigs for pinned beads. This ensures the propulsion principle works
regardless of which directory the agent starts in.
The scan uses routes.jsonl to find all rig beads directories.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When slinging or hooking work to mayor/deacon, the pin now lands in Town
beads (~/.beads/) instead of rig beads. This ensures gt mol status finds
the pinned work when run from ~/gt.
The issue was that town-level roles operate from the town root, so their
hooks should be discoverable from there.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The template incorrectly used --assignee flag on bd mol wisp which doesn't
exist. Changed to two-step pattern matching deacon template.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Changed polecat branch model to use unique timestamped branches
(polecat/<name>-<timestamp>) instead of reusing persistent branches.
This prevents JSONL drift issues where stale polecat branches don't
have recently created beads.
Changes:
- Add(): create unique branch, simplified (no reuse logic)
- Recreate(): create fresh branch, old ones left for GC
- loadFromBeads(): read actual branch from git worktree
- CleanupStaleBranches(): remove orphaned polecat branches
- ListBranches(pattern): new git helper for branch enumeration
- gt polecat gc: new command to clean up stale branches
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When run in a non-crew session (Mayor, Witness, Refinery, Deacon),
gt crew next/prev now returns success (exit 0) instead of an error.
These sessions don't have cycle groups, so pressing C-b n should
simply do nothing rather than fail with an error message.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
These methods called `bd pin` and `bd unpin` which have been removed.
Neither method was ever called - gt uses `bd update --status=pinned` instead.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add --all flag to restart all running crew sessions
- Add --dry-run flag to preview without restarting
- Add --rig filter to target specific rig
- Extract restartCrewSession helper for reuse
🤝 Filed gt-1kljv for adding tests
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Theme is configuration (from settings/config.json), not runtime state.
Loading theme from state file was causing rig config to be ignored,
which is why the beads rig kept using mad-max names instead of minerals.
Crew workers are human-managed and should never be auto-killed by
gt doctor --fix. This adds defense-in-depth protection:
1. OrphanSessionCheck.Fix() now skips any session matching the
gt-<rig>-crew-<name> pattern
2. OrphanProcessCheck.Fix() now checks if a process has a crew
session pane as an ancestor before killing it
Even if detection fails (like the pgrep bug we just fixed), crew
sessions and their processes will be protected.
Generated with Claude Code
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The orphan-processes check was incorrectly killing active crew sessions
because pgrep -x tmux does not reliably find the tmux server on macOS.
Root cause:
- pgrep -x tmux was finding tmux attach-session processes but missing
the actual tmux server process
- Claude processes running in tmux panes were incorrectly flagged as
orphaned because their parent (tmux server) was not in the allowed list
Fixes:
1. Use ps + awk instead of pgrep to find tmux processes more reliably
2. Exclude Claude.app desktop processes from orphan detection (they are
not Gas Town CLI processes)
Closes: gt-ronyn
Generated with Claude Code
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When slinging a bead that has the template label (a proto), and --var
flags are provided, automatically call bd --no-daemon pour to
instantiate the proto with variable substitution before slinging.
This enables the seamless workflow:
gt sling mol-release beads --var version=0.38.0
Instead of the manual two-step:
bd --no-daemon pour mol-release --var version=0.38.0 --json
gt sling <resulting-mol-id> beads
Also adds a warning when slinging a proto without --var, since
{{variables}} will not be substituted.
Generated with Claude Code
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Adds comprehensive "Theory of Operation" section explaining WHY agents
must execute work immediately when found on their hook. Uses steam engine
metaphor to convey that this is physics, not politeness.
Each role gets a tailored metaphor:
- Mayor: main drive shaft
- Crew/Polecat: piston
- Deacon: flywheel
- Witness: pressure gauge
- Refinery: exhaust valve
The section explains the failure mode (agent waits, human is AFK,
Gas Town stops) and reinforces the propulsion principle.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- gt unsling: clear your own hook
- gt unsling <bead>: only unsling if that bead is hooked
- gt unsling <target>: clear another agent's hook
- gt unsling <bead> <target>: unsling specific bead from agent
- gt unhook: alias for gt unsling
Symmetric with gt sling/hook commands.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Beads CLI restructured: pour and wisp are now subcommands of mol.
- bd pour → bd mol pour
- bd wisp → bd mol wisp
Updated all documentation, templates, and code to use new command structure.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Delete all .formula.json files (TOML versions exist)
- Update sling.go comment to say TOML only
Note: Doc updates were prepared but those docs were deleted
upstream in the recent refactoring.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Fix crew next/prev: Pass session name via key binding to avoid run-shell context issue
- Add TouchTownActivity() for town-level activity signaling
- Implement daemon exponential backoff based on activity.json:
- 0-5 min idle → 5 min heartbeat
- 5-15 min idle → 10 min heartbeat
- 15-45 min idle → 30 min heartbeat
- 45+ min idle → 60 min heartbeat (max)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Previously, gt doctor --fix would kill workers whose spawning process had
exited, even though the Claude session was still running in tmux.
Now both identity_check.go and CleanStaleLocks check if the tmux session
exists before declaring a lock stale. A lock is only truly stale if BOTH
the PID is dead AND the session does not exist.
Also added ListSessionIDs() to tmux package to handle locks that store
session IDs (%N or $N format) instead of session names.
Adds CloneDivergenceCheck that detects when git clones have drifted
significantly behind origin/main:
- >10 commits behind: WARNING
- >50 commits behind: ERROR (EMERGENCY)
Only checks clones on main branch, since off-main clones are already
caught by BranchCheck. This distinguishes from beads-sync divergence
which is expected behavior.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add internal/beads/routes.go with helpers to manage routes.jsonl
- Update gt rig add to auto-append routes for new rigs
- Add prefix-conflict check to gt doctor
bd already has prefix routing via routes.jsonl - this wires up
Gas Town to maintain routes when rigs are added and detect conflicts.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
bd defaults to $BD_ACTOR → git config user.name → $USER, which is
more accurate for system tools like doctor and rig init.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
gt start now reads rig settings (settings/config.json) and auto-starts
configured crew members. The crew.startup field supports:
- Single name: "dave"
- Multiple names: "max and joe" or "max, joe"
- All crew: "all"
- None: "none" or empty
Configured:
- beads: dave
- gastown: max and joe
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add CreatedBy field to Issue struct (matches beads GH#748)
- Add Actor field to CreateOptions, pass --actor to bd create
- Add ActorString() method to RoleInfo for identity formatting
- Update all beads.Create() callers to pass Actor
- Update direct bd create exec calls with --actor:
- mail/router.go: uses sender identity
- patrol_helpers.go: uses role name
- doctor/patrol_check.go: uses "gt-doctor"
- rig/manager.go: uses "gt-rig-init"
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The crash logging fallback assumed users might have their town at
~/gastown, which only works if the gastown rig is installed. Now
only tries ~/gt as the conventional town root.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When GT_ROLE is set to a simple value like "crew", the role detection
was not reading GT_RIG and GT_CREW to get the full identity. This caused
gt mol status to show "/crew/" instead of "gastown/crew/max".
Now GetRoleWithContext checks these additional env vars when the rig
or polecat fields are empty after parsing the role string.
Generated with Claude Code
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Messages now show [from mayor] or [from beads/crew/dave] etc.
so recipients know who sent the nudge.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Detect crew/ prefix in polecatName and use crewSessionName() instead
of polecat session manager. This produces correct session names like
gt-beads-crew-dave instead of gt-beads-p-crew/dave.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add guard that checks if bead is already pinned before slinging.
Shows current assignee and requires --force to override.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
