When slinging to an existing polecat with --naked flag, the code was still
attempting to look up the tmux pane, which fails for terminated polecats.
Now resolveTargetAgent accepts a skipPane parameter that bypasses the tmux
pane and working directory lookup when true.
This allows work to be slung to terminated polecats that will be restarted
manually later.
Also updated unsling to skip pane lookup since it only needs the agent ID.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
CLAUDE.md is clone-specific (contains crew member name and working
directory) and should not be tracked in git.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
CLAUDE.md contains clone-specific content (crew member name, working
directory) and should not be tracked in git. Each clone regenerates
its CLAUDE.md locally when created via `gt crew add`.
This prevents crew workers from getting incorrect identity context
when pulling changes made by other crew members.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The crew CLAUDE.md files were incorrectly using polecat template
content. This regenerates them with the proper crew template.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The main polecat spawn fix was already in place (ec29ca07), but the naked
mode help text was telling users to run `claude` without the flag. Updated
to show the correct command with --dangerously-skip-permissions.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Implements the primary wake mechanism for patrol agents:
- Subscribes to bd activity --follow as a background process
- Returns immediately when any line of output is received
- Timeout with optional exponential backoff as safety net
- JSON output mode for scripting
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Before: gt crew at only looked for tmux sessions with the specific naming
convention gt-<rig>-crew-<name>. If the user started Claude manually or
via a different mechanism, it would create a duplicate session.
After: Before creating a new session, check if any existing tmux session
has Claude running in the crews directory. If found, attach to that
session instead of creating a new one.
Changes:
- Add FindSessionByWorkDir() to internal/tmux/tmux.go to search sessions
by working directory, optionally filtering for Claude (node) running
- Update runCrewAt() to check for existing sessions before creating new
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When starting a polecat session with --issue flag, the issue is now
properly hooked to the polecat via bd update --status=hooked. This
ensures 'gt mol status' shows the assigned work when the session starts.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Two bugs fixed in loadTasksFromBeads():
1. JSON field name mismatch: code parsed dependencies but bd show
returns dependents - meant worker discovery always failed
2. Missing Assignee field: even if field name was correct, assignees
were not being extracted from the bd show output
Also added hooked status to TaskInProgress mapping since workers
with hooked beads are actively working.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Adds a new 'gt commit' command that wraps git commit and automatically
injects agent identity trailers into commit messages:
- Executed-By: agent identity (e.g., beads/crew/dave)
- Rig: the rig name
- Role: crew, polecat, witness, etc.
- Molecule: pinned molecule ID if any
This enables forensic analysis and audit trails for agent-mediated commits.
Supports common git commit flags: -a, --amend, --no-edit, --allow-empty.
Use --no-trailers to skip adding identity trailers.
- Add LoadSwarm method for loading swarms from beads
- Rewrite tests to use LoadSwarm, GetSwarm, GetReadyTasks, IsComplete
- Remove tests for deprecated Create, Start, UpdateState methods
- Keep E2E lifecycle documentation test
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Change daemon from wake-focused to recovery-focused:
Before: Daemon pokes agents every 5-60min as primary wake
After: Daemon only checks for edge cases that feed-wake cannot handle
Recovery checks:
- Dead sessions that need restart (ensureDeaconRunning, ensureWitnessesRunning)
- Stale agents that crashed without updating state (checkStaleAgents)
- GUPP violations: agents with work-on-hook not progressing (checkGUPPViolations)
- Orphaned work: work assigned to dead agents (checkOrphanedWork)
Removed:
- pokeDeacon() - no longer sending HEARTBEAT messages
- pokeWitness()/pokeWitnesses() - no longer sending HEARTBEAT messages
- MOTD message arrays - only used by poke functions
Normal agent wake is now handled by feed subscription (bd activity --follow).
The daemon is the safety net for edge cases, not the primary propulsion.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Adds Type and Backoff fields to MoleculeStep for patrol agents to implement
cost-saving await-signal patterns:
- Type field: "task" (default), "wait" (await-signal), etc.
- BackoffConfig: base interval, multiplier, max cap
- Parsing for "Type:" and "Backoff:" lines in step definitions
- Comprehensive tests for new parsing functionality
Step definition format:
## Step: await-signal
Type: wait
Backoff: base=30s, multiplier=2, max=10m
Agents interpret these declaratively, implementing backoff behavior at runtime.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add special handling for 'deacon' target in gt nudge command
- Maps 'deacon' to gt-deacon session, gracefully handles if not running
- Add gt nudge deacon session-started to SessionStart hooks
- Updated settings-autonomous.json, settings-interactive.json, and
ensurePatrolHooks() template
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Adds first-class support for the human overseer in Gas Town mail:
- New OverseerConfig in internal/config/overseer.go with identity
detection (git config, gh cli, environment)
- Overseer detected/saved on town install (mayor/overseer.json)
- Simplified detectSender(): GT_ROLE set = agent, else = overseer
- New overseer address alongside mayor/ and deacon/
- Added --cc flag to mail send for CC recipients
- Inbox now includes CC'd messages via label query
- gt status shows overseer identity and unread mail count
- New gt whoami command shows current mail identity
Generated with Claude Code
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The swarm Manager was maintaining an in-memory map of swarms that was
never persisted and duplicated state from beads. This caused stale state
after restarts and confusion about source of truth.
Changes:
- Remove swarms map from Manager (now stateless)
- Add LoadSwarm() that queries beads for swarm state
- Refactor all methods to use LoadSwarm() instead of in-memory lookup
- Discover workers from assigned tasks in beads
- Remove obsolete unit tests that tested in-memory behavior
- Keep type/state tests that do not need beads
The E2E test (gt-kc7yj.4) now covers the beads integration.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Changed icon from 📌 to 🪝 for hooked beads
- Changed label from "Pinned" to "Hooked"
- Added 🚀 AUTONOMOUS MODE banner when work is on hook
- Clarified that hooked bead triggers autonomous work even without molecule
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add "The Capability Ledger" section to all 6 role templates explaining:
- Work visibility and reputation
- Redemption through consistent quality
- Every completion as evidence of autonomous execution at scale
- Work history as growing portfolio/CV
Also fix hooked vs pinned terminology confusion:
- "Hooked" = work assigned to you (triggers autonomous mode)
- "Pinned" = permanent reference beads
- Add clarifying note to all startup behavior sections
- Update code examples to use --status=hooked
Update prime.go AUTONOMOUS MODE output with ledger framing.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When gt crew remove runs, it now closes the associated agent bead
(gt-<rig>-crew-<name>) with reason "Crew workspace removed".
This prevents orphaned agent beads in the system after crew removal.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Update mol-witness-patrol formula to:
1. Close the epic (swarm molecule) when all children are closed
2. Close the swarm tracking wisp
3. Notify Mayor with SWARM_COMPLETE mail
Activity feed events are generated automatically by bd close commands.
(gt-kc7yj.3)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- checkSlungWork now queries for status=hooked instead of status=pinned
- This makes AUTONOMOUS WORK MODE fire correctly when work is slung via gt hook
- Added "physics not politeness" language to AUTONOMOUS MODE output
- Added full Theory of Operation section to refinery.md.tmpl for consistency
The bug was: gt hook sets status=hooked but gt prime was querying for
status=pinned, so AUTONOMOUS MODE never triggered for hooked work.
(gt-kc7yj.3)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
