Complete the statusline optimization by adding cache writes to all
output functions. The existing patch added cache functions and cache
reads, but never wrote to the cache.
Changes:
- Add early-return for detached sessions (return static "○ |")
- Add cache read check for attached sessions
- Add setStatusLineCache() calls in all 5 output functions:
- runWorkerStatusLine
- runMayorStatusLine
- runDeaconStatusLine
- runWitnessStatusLine
- runRefineryStatusLine
This should reduce Dolt CPU from ~70% to ~20% when agents are idle,
as tmux status lines will use cached results instead of spawning
beads queries every 5 seconds.
Testing: Run `nix switch` then monitor Dolt CPU with `top`
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add title-based lookup for hq- prefixed beads (uses title as address if contains "/")
- Add rig-specific prefix handling to parse IDs like j-java-crew-americano → java/crew/americano
- Handles crew, polecat, witness, refinery role patterns
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Fix mail router normalization in validateRecipient
- Fix agentBeadToAddress to use title field for hq- prefixed beads
- Fix crew/polecat home paths (remove incorrect /rig suffix)
- Fix town root detection (RoleUnknown instead of RoleMayor)
- Fix copyDir symlink handling
- Pin to gastown commit 177094a matching jt flake
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Fix mail router normalization in validateRecipient
- Fix agentBeadToAddress to use title field for hq- prefixed beads
- Fix crew/polecat home paths (remove incorrect /rig suffix)
- Fix town root detection (RoleUnknown instead of RoleMayor)
- Fix copyDir symlink handling
- Pin to gastown commit 177094a matching jt flake
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The gastown build now requires BuiltProperly=1 to be set via ldflags,
otherwise gt errors with "This binary was built with 'go build' directly".
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Files copied from the nix store inherit read-only permissions, causing
subsequent home-manager activations to fail with "Permission denied".
Add rm -f before copy and chmod u+w after copy for all plugin files:
- humanlayer commands and agents
- local commands and skills
- micro-skills
- beads formulas
Executed-By: nixos_configs/crew/hermione
Rig: nixos_configs
Role: crew
- Add gastown flake input (non-flake, source only)
- Build gastown package using buildGoModule in development role
- Configure renovate for daily updates of gastown and beads
- Binary: gt (Gas Town CLI by Steve Yegge)
Add periodic timer (every 5 min) that runs `bd gate check --type=timer`
across all beads workspaces with running daemons.
This enables self-scheduling molecules - workflows that can set timer
gates to schedule their own future execution (e.g., watchers that scan
Slack every 8 hours).
The timer:
- Only runs on Linux (uses systemd user services)
- Discovers workspaces via `bd daemon list`
- Silently skips if no beads daemons are running
Related: nixos-configs-1rk.2 (Time beads scheduling mechanism)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add skills/ directory for local Claude skills
- Create parallel_beads.md skill that orchestrates:
- Phase 1: Multi-select bead selection from bd ready
- Phase 2: Parallel subagents for implementation (worktree, implement, commit, PR)
- Phase 3: Parallel review subagents
- Phase 4: Cleanup and summary
- Update default.nix to install local skills alongside humanlayer plugins
- Support both gh (GitHub) and tea (Gitea/Forgejo) based on origin URL
Adds a custom Nix package for perles, a terminal user interface for the
Beads issue tracking system with BQL query language support.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Replace complex marketplace plugin installation with simple 'bd setup claude'
which installs hooks into ~/.claude/settings.json. This is the recommended
approach for Claude Code (CLI + hooks) vs the MCP server approach.
Development environments now use standard nixpkgs claude-code by default.
Work environments override with custom binary distribution to bypass
corporate npm registry restrictions via Google Cloud Storage.
Add allowArbitraryClaudeCodeModelSelection option to development role
to control whether model specifications are preserved in humanlayer
commands and agents. Defaults to false (strip models) for backward
compatibility. Enable on home-darwin-work to preserve model specs
for opus/sonnet selection.
Create custom Nix package for Claude Code CLI that fetches directly from
Anthropic's Google Cloud Storage distribution instead of npm registry,
working around Block's Cloudflare Teams dependency confusion protection.
- Add claude-cli package with platform-specific binaries
- Include comprehensive README with update instructions
- Enable development role on work machine
- Switch from unstable.claude-code to custom.claude-cli
- Add google-cloud-sdk to work machine packages
Removes model specification to better support Claude Code Pro plan.
Because of this, I also re-removed this role from home-darwin-work as we
can take advantage of more powerful models in that context. We will just
need to install the plugins external from nix. Eventually I can turn
this into a config option of course.
Also made sure we are installing the agents from the plugin in addition
to the commands.
