c0fc16e25c
The repo is maintained by agents from multiple vendors, so agent instructions move out of vendor-specific locations into shared, git-tracked markdown: - AGENTS.md is the new entry point for any harness; CLAUDE.md becomes a symlink to it (Claude Code keeps working unchanged). - Skill bodies relocate to agent/: VERIFICATION.md (the enforcement rules, ex .claude/skills/nomarchy), DELEGATION.md (capability tiers light/standard/frontier, scout/runner role contracts, token economy, parallel fan-out — consolidates the CLAUDE.md model table, LOOP.md's economy section, and skill §6.5 into one place; vendor model names survive only in the per-harness mapping table), THEME-DESIGN.md (ex .claude/skills/theme, which previously lacked frontmatter). - .claude/ shrinks to a thin Claude Code adapter: settings, subagent defs, and skill shims that route into agent/. - Maps updated: agent/README.md (instructions vs state vs adapters), docs/README.md, README.md layout tree. Cleanup: stale old_distro .gitignore entry dropped (dir long gone); local result*/__pycache__ artifacts removed; JOURNAL.md rotated (29 recent entries kept, 120 older moved to agent/JOURNAL-ARCHIVE.md, rotation rule documented in the header). Verification: V0 (docs/meta only) — nix flake check --no-build exit 0 with the tracked symlink; grep sweep confirms no vendor model names outside DELEGATION.md's mapping table. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2.7 KiB
Nomarchy — agent entry point
Nomarchy is a NixOS-based distro: rock-stable, fully reproducible, themed from one JSON, configured through a menu that writes into the user's own flake checkout. Read the README for the architecture.
This file is the entry point for any AI coding agent, whatever the
vendor or harness. Everything agents need lives in vendor-neutral,
git-tracked markdown (agent/, docs/); harness-specific config is a
thin adapter (see bottom) and never holds shared content.
Where things live (read this first)
| Path | Role |
|---|---|
agent/ |
Agent instructions + loop state — only executable queue is BACKLOG.md |
agent/README.md |
Map of the agent files |
docs/VISION.md |
Product themes toward v1.0 (not a queue) |
docs/ROADMAP.md |
Design history + shipped log |
docs/README.md |
Full documentation map |
.claude/ |
Claude Code adapter only (permissions, subagents, skill shims) — not the backlog |
If you're here to work autonomously (the loop)
Follow agent/LOOP.md — one iteration: orient → pick one BACKLOG
item → work → verify → commit+push on main → record. All loop state is
git-tracked in agent/.
Rules that apply to every session, loop or not
- Honesty rule (
agent/VERIFICATION.md,docs/TESTING.md): for visual/interactive changes, evaluation is not rendering — state the tier you reached. Cheap first:nix flake check --no-build. - Conventions (
agent/CONVENTIONS.md): in-flake state, menu placement, Waybar parity with whole-swaps, toggle-vs-package discipline, template as SoT for opt-in comments (templates/downstream). - Git: direct commits on
main, pushed;v1is human-only — never touch it. Nevernix flake updateunless the task is a lock bump. No formatter — match aligned hand-formatting. - Layout:
hosts/machine ·modules/distro ·themes/data ·pkgs/code ·tools/maintainer ·agent/loop ·docs/human docs.
Delegation & escalation
Push mechanical work down to cheaper models; keep judgment on the
strongest model available. Capability tiers, standing roles (scout /
runner), and per-harness model mappings: agent/DELEGATION.md.
Work above your tier? Return it — don't burn tokens.
Harness adapters
One directory per harness, holding only what that harness requires in
its own format (permissions, subagent/skill registration). Adapters point
into agent/; they never carry policy, queue items, or vision text.
| Harness | Adapter |
|---|---|
| Claude Code | .claude/ — CLAUDE.md is a symlink to this file |
| others | add a sibling dir + a row here; keep it a thin shim |