Share repository guidance across Codex and Claude Code

README.md

@@ -112,7 +112,7 @@ bazilion completion bash|zsh|fish          # print a shell completion script
 - **Pi agent engine** — Bazilion is based on [pi-coding-agent](https://www.npmjs.com/package/@earendil-works/pi-coding-agent). Pi runs each turn, stores the canonical JSONL transcript under the agent's `sessions/` directory, handles replay and compaction, executes the provider/tool loop, and supplies the coding tools. Bazilion contributes the multi-agent shell around that engine: profiles, groups, USER.md, memory, mailbox, scheduler, browser/MCP integrations, and clients.
 - **Group** — a collaboration context: one filesystem root, one USER.md, one roster, one shared memory. Every agent belongs to exactly one group, chosen at spawn time. The agent's coding tools (`read`/`bash`/`edit`/`write`/`grep`/`find`/`ls`, supplied by [pi-coding-agent](https://www.npmjs.com/package/@earendil-works/pi-coding-agent)) are rooted at the group directory. USER.md is read-only to agents — edit it via `bazilion group user-md set` or the web UI. First-run setup seeds a `default` group at `~/.bazilion/groups/default/`. Groups always live under `~/.bazilion/groups/<slug>/`; pass `--link <existing-path>` to `bazilion group add` to materialize the slot as a symlink to your existing project tree instead of as a fresh directory.
 - **Agent** — an instance spawned from a profile into a group. Has a private home (`~/.bazilion/agents/<id>/` — its copy of the templates, plus pi's append-only session JSONL under `sessions/`) reachable via the `home_*` tools, and one group membership reachable via the coding tools. UUIDs as ids.
-- **Skill** — a directory under `~/.bazilion/skills/<name>/` with a `SKILL.md` (standard OpenClaw / Anthropic agent-skill format). Imported via `bazilion skill import --from openclaw` (or any path). The body is injected into the system prompt of every agent the skill is attached to; helper scripts shipped alongside the markdown are invoked by the agent via its generic `bash` tool (no framework-level entrypoint and no trust gate — see CLAUDE.md for why we removed both).
+- **Skill** — a directory under `~/.bazilion/skills/<name>/` with a `SKILL.md` (standard OpenClaw / Anthropic agent-skill format). Imported via `bazilion skill import --from openclaw` (or any path). The body is injected into the system prompt of every agent the skill is attached to; helper scripts shipped alongside the markdown are invoked by the agent via its generic `bash` tool (no framework-level entrypoint and no trust gate — see AGENTS.md for why we removed both).
 - **Memory** — **group-shared** BM25 index rooted at `<groupPath>/memory/`. Every agent in the group reads + writes the same store. The current backend is `qmdBackend` (BM25 over markdown via [@tobilu/qmd](https://github.com/tobi/qmd)). Use it for project knowledge — codebase notes, decisions, things the user told you about the work; for personal notes about an agent (preferences, persona quirks), use `home_write` on `IDENTITY.md` instead.
 - **Mailbox** — `messages` table. Agents talk to each other via `send_message` / `read_inbox` / `wait_for_reply` tools, via `bazilion send` from the CLI, or from outside the loop: `bazilion inbox list <agent> [--unread]`, `bazilion inbox show <id>`, `bazilion inbox read <id>`, or the web UI at `/agents/<id>/inbox`. The worker delegates these tool calls to the daemon over Node IPC — workers don't hold their own SQLite handle.
 - **Trigger** — a heartbeat (interval in seconds) or cron expression that periodically wakes an agent with a stored message. An in-process scheduler ticks every 5 s (overridable via `BAZILION_SCHEDULER_TICK_MS`; disable with `BAZILION_SCHEDULER=off`) and fires due triggers through the same code path as user chat. Example: `bazilion trigger add <agent> --every 300 --message "check your inbox"`.

apps/web/src/lib/wire-constants.ts

@@ -4,7 +4,7 @@
 // (`apps/daemon/src/core`) and re-exported through `@bazilion/api-types`. We
 // re-declare them here rather than reaching into daemon source so the web
 // app stays decoupled from Node-only deps (the architecture mandate is
-// "apps/web never reaches into daemon source" — see CLAUDE.md). Daemon-side
+// "apps/web never reaches into daemon source" — see AGENTS.md). Daemon-side
 // validation re-checks every wire string against the source of truth, so
 // a drift here surfaces as a 4xx, never as silent corruption.
 

docs/architecture.md

@@ -134,7 +134,7 @@ Uses Node 22's built-in `node:sqlite` (`DatabaseSync`). No `better-sqlite3`, no
 | `agent_skills` | `(agent_id, skill_name)` — per-agent skill attachments. Cascade on agent delete. |
 | `agent_triggers` | Per-agent wake-ups (`kind='interval' \| 'cron'`, `interval_sec`, `cron_expr`, `message`, `last_fired_at`, `enabled`). |
 | `messages` | Inter-agent mailbox (`id`, `from_agent_id`, `to_agent_id`, `payload`, `reply_to`, `read_at`). FKs do **not** cascade; `agent/delete.ts` nulls inbound `reply_to` and purges rows manually before removing the agent. |
-| `skill_meta` | Import provenance — `(name PK, source, imported_at)`. No trust column — see "Skill model" in CLAUDE.md. |
+| `skill_meta` | Import provenance — `(name PK, source, imported_at)`. No trust column — see "Skill model" in AGENTS.md. |
 | `web_tokens` | Per-token access records (hashed). The `bootstrap` row's plaintext lives in `auth.json`; revoking it returns 409 (would lock the operator out). |
 | `provider_models` | Curated model list per provider (what shows up in dropdowns). |
 | `provider_state` | Per-provider enabled flag. |
@@ -389,7 +389,7 @@ Cancellation: `POST /api/agents/:id/cancel` → `cancelAgent(id)` → the stored
 
 ## 4b. `apps/web` — the browser UI (TanStack Start)
 
-React 19 + Vite 7 + Tailwind v4 + shadcn/ui. File-based routes via `@tanstack/react-router`; `src/router.tsx` exports `getRouter`. Server fns use `.inputValidator()`. Bound to `127.0.0.1:4322` by default (`WEB_HOST` / `WEB_PORT`).
+React 19 + Vite 8 + Tailwind v4 + shadcn/ui. File-based routes via `@tanstack/react-router`; `src/router.tsx` exports `getRouter`. Server fns use `.inputValidator()`. Bound to `127.0.0.1:4322` by default (`WEB_HOST` / `WEB_PORT`).
 
 **Daemon-only client** — `apps/web` never reaches into daemon source; the only daemon-facing types it imports are wire shapes from `@bazilion/api-types`. Every loader is a `createServerFn` handler that calls `apps/web/src/lib/daemon-client.ts`, which reads the request's `bz_token` cookie via `getCookie` from `@tanstack/react-start/server` and forwards it as `Authorization: Bearer …` to the daemon (`http://127.0.0.1:4321`, overridable via `BAZILION_DAEMON`).
 

docs/openclaw-reference.md

@@ -1,6 +1,6 @@
 # OpenClaw — reference for Bazilion contributors
 
-Bazilion's README and `CLAUDE.md` call the project "OpenClaw-inspired", and the daemon literally reads `~/.openclaw/skills/` when you run `bazilion skill import --from openclaw`. This doc captures what upstream OpenClaw actually is, so contributors know what we borrowed, what we deliberately didn't, and where the format we accept comes from.
+Bazilion's README and `AGENTS.md` call the project "OpenClaw-inspired", and the daemon literally reads `~/.openclaw/skills/` when you run `bazilion skill import --from openclaw`. This doc captures what upstream OpenClaw actually is, so contributors know what we borrowed, what we deliberately didn't, and where the format we accept comes from.
 
 Snapshot date: 2026-05-17. Source: `docs.openclaw.ai` + `github.com/openclaw/openclaw` `main`. Re-fetch before relying on exact counts.
 
@@ -179,7 +179,7 @@ Official registry: `github.com/openclaw/clawhub`. ~13,729 third-party skills as
 
 **Borrowed:**
 
-- The **prompt-only skill model.** `apps/daemon/src/core/skills/` reads `SKILL.md` and injects the body into the system prompt of every attached agent. Helper scripts run via the agent's generic `bash` tool. See `CLAUDE.md` → "OpenClaw skill model: prompt-only".
+- The **prompt-only skill model.** `apps/daemon/src/core/skills/` reads `SKILL.md` and injects the body into the system prompt of every attached agent. Helper scripts run via the agent's generic `bash` tool. See `AGENTS.md` → "OpenClaw skill model: prompt-only".
 - The **`SKILL.md` + YAML frontmatter format.** Upstream skills "drop in unchanged" per the README — `bazilion skill import --from openclaw` resolves to `~/.openclaw/skills/` (see `apps/daemon/src/routes/skills.ts`).
 - The general idea of a per-user state root with a workspace concept. Bazilion's `~/.bazilion/groups/<slug>/` is roughly the analogue of OpenClaw's `<workspace>/`.
 
@@ -192,7 +192,7 @@ Official registry: `github.com/openclaw/clawhub`. ~13,729 third-party skills as
 - The **plugin model** (28 hooks, TypeScript in-process modules, `openclaw.plugin.json`). Bazilion has no plugin SDK — all extensibility goes through skills + native daemon code.
 - **Channels.** Upstream's first-class WhatsApp/Telegram/iMessage/Signal transports have no Bazilion equivalent; we expose HTTP only.
 - **ClawHub.** No registry integration.
-- **JSON5 `openclaw.json`** config surface. Bazilion stores config in the daemon's SQLite (`config` table) + secrets in the encrypted `secrets` table (see `CLAUDE.md` → "Secrets and config").
+- **JSON5 `openclaw.json`** config surface. Bazilion stores config in the daemon's SQLite (`config` table) + secrets in the encrypted `secrets` table (see `AGENTS.md` → "Secrets and config").
 - The richer skill frontmatter (`user-invocable`, `command-dispatch`, `disable-model-invocation`, install recipes). Bazilion's skill loader only consumes `name` + `description` + body today.
 
 ## 7. Authoritative sources