bajaclaw 0.10.1

package/skills/setup-memory-sync/SKILL.md

@@ -0,0 +1,45 @@
+---
+name: setup-memory-sync
+description: Enable two-way memory sync between BajaClaw and ~/.claude/memory/
+version: 0.1.0
+tools: [Bash, Read, Edit, Write]
+triggers: ["memory sync", "share memory", "sync memory", "connect memory", "memory compat"]
+effort: low
+---
+
+## When to use
+User wants facts BajaClaw learns across cycles to show up in desktop CLI
+sessions (and vice versa), without manual copy/paste.
+
+## Quick reference
+- Direction: bidirectional, per-profile, opt-in.
+- BajaClaw side: FTS5 `memories` table in `bajaclaw.db`.
+- Desktop side: plain markdown files under `~/.claude/memory/`.
+- Implementation: `src/memory/claude-compat.ts`.
+
+## Procedure
+1. Edit `~/.bajaclaw/profiles/<profile>/config.json`. Add or update:
+   ```json
+   { "memorySync": true }
+   ```
+2. On the next cycle, BajaClaw will:
+   - Ingest any new/modified `*.md` files under `~/.claude/memory/` into
+     its FTS table with `source=claude-code`.
+   - Write a digest to `~/.claude/memory/bajaclaw-<profile>.md` so desktop
+     sessions see what BajaClaw has been learning.
+3. Verify:
+   - `bajaclaw dashboard <profile>` — Memories panel will include entries
+     with kind `claude-code`.
+   - Check `~/.claude/memory/bajaclaw-<profile>.md` for the digest file.
+
+## Pitfalls
+- Sync is deliberate, not automatic. Only profiles with `memorySync: true`
+  participate.
+- The digest file overwrites on each sync — it's a snapshot, not an
+  append-log. Don't hand-edit it.
+- Large `~/.claude/memory/` directories slow down cycles. If you have
+  thousands of files there, consider pruning before enabling sync.
+
+## Verification
+- Post-cycle: `bajaclaw dashboard` shows memories with `kind=claude-code`
+- `~/.claude/memory/bajaclaw-<profile>.md` exists and contains recent memories

package/skills/setup-profile/SKILL.md

@@ -0,0 +1,44 @@
+---
+name: setup-profile
+description: Create an additional named BajaClaw profile beyond the default
+version: 0.1.0
+tools: [Bash, Read, Write]
+triggers: ["new profile", "add profile", "another agent", "second agent", "create profile", "multiple agents"]
+effort: low
+---
+
+## When to use
+User wants more than one agent — e.g. one for research, one for inbox
+triage, one for coding. Each profile gets its own DB, skills, schedule,
+logs, and agent descriptor.
+
+## Quick reference
+- Location: `~/.bajaclaw/profiles/<name>/`
+- Agent descriptor: `~/.claude/agents/<name>/<name>.md` (auto-written)
+- Templates: `outreach | research | support | social | code | custom`
+- Switch: positional arg or `BAJACLAW_PROFILE` env var.
+
+## Procedure
+1. Ask the user what this agent is for — that picks the template.
+2. Run: `bajaclaw init <name> --template <tpl>`
+   - Example: `bajaclaw init researcher --template research`
+3. Optional: set model + effort at init:
+   `bajaclaw init researcher --template research --model claude-opus-4-7 --effort high`
+4. Edit `~/.bajaclaw/profiles/<name>/AGENT.md` and `SOUL.md` to tailor the
+   agent's operating rules and identity.
+5. Edit `~/.bajaclaw/profiles/<name>/HEARTBEAT.md` with schedule lines.
+6. First run: `bajaclaw start <name> --dry-run` to verify the assembled
+   prompt, then `bajaclaw start <name>` for a live cycle.
+
+## Pitfalls
+- Profiles don't share memory unless you enable `memorySync` (via
+  `~/.claude/memory/` digests). They're independent by design.
+- `bajaclaw profile delete <name> --yes` removes the profile directory.
+  Not recoverable. Back up `bajaclaw.db` first if you want history.
+- Two profiles can't run the dashboard on the same port. Pick different
+  `dashboardPort` values in each `config.json`.
+
+## Verification
+- `bajaclaw profile list` shows the new profile
+- `bajaclaw status <name>` returns stats
+- `ls ~/.claude/agents/<name>/` shows the `<name>.md` descriptor

package/skills/setup-self-update/SKILL.md

@@ -0,0 +1,45 @@
+---
+name: setup-self-update
+description: Check for and install a new BajaClaw version, or configure the update channel
+version: 0.1.0
+tools: [Bash, Read]
+triggers: ["update bajaclaw", "upgrade bajaclaw", "new version", "auto update", "check updates", "silence update notice"]
+effort: low
+---
+
+## When to use
+User wants to update BajaClaw, see what version they're on, or silence the
+update-available notice.
+
+## Quick reference
+- Check + install: `bajaclaw update`
+- Current version: `bajaclaw --version`
+- Cache: `~/.bajaclaw/.update-check.json` (24h TTL)
+- Silence notice: `BAJACLAW_NO_UPDATE_NOTICE=1`
+- Channels: npm registry (primary) + optional GitHub raw fallback
+  configured in `package.json` → `bajaclaw.updateUrl`.
+
+## Procedure
+1. Check delta without installing: `bajaclaw update --check`
+2. Apply: `bajaclaw update --yes`
+   - Installed via `npm install -g bajaclaw`: runs
+     `npm install -g bajaclaw@latest`.
+   - Installed from a git clone: runs
+     `git pull && npm install && npm run build`.
+   - Installed via `npm install -g github:backyarddd/BajaClaw`
+     (bleeding-edge track): `bajaclaw update` still reinstalls from
+     npm registry latest. Re-run the github install to stay on HEAD.
+3. Silence the post-command notice (e.g. in scripts):
+   export `BAJACLAW_NO_UPDATE_NOTICE=1` in the shell config.
+4. To force a fresh check (bypassing the 24h cache):
+   `rm ~/.bajaclaw/.update-check.json && bajaclaw update --check`
+
+## Pitfalls
+- If `npm install -g` requires sudo on their system, the update will fail
+  silently. Ask them to retry with their node-version manager active or
+  fix the npm prefix.
+- Network unavailable → silent fail, but the CLI still works. Retry later.
+
+## Verification
+- `bajaclaw --version` reports the new version.
+- Update notice disappears after a successful apply.

package/skills/setup-subagent/SKILL.md

@@ -0,0 +1,102 @@
+---
+name: setup-subagent
+description: Create a specialized sub-agent with its own tools and permissions, owned by an orchestrator
+version: 0.1.0
+tools: [Bash, Read, Write, Edit]
+triggers: ["create subagent", "setup subagent", "new subagent", "add helper agent", "email agent", "mail agent", "calendar agent", "scope an agent"]
+effort: medium
+---
+
+## When to use
+The user wants a specialized helper agent with its own permissions that the
+main agent can delegate to. Common cases: a "mail" agent with email MCP that
+the main agent routes inbox tasks to; a "finance" agent that has access to
+banking/budgeting tools the main doesn't.
+
+## Quick reference
+- Sub-agents are separate BajaClaw profiles with a `parent:` pointer.
+- Parent's `config.json` gets a `subAgents: [...]` list.
+- Each sub-agent has its own tools, memory, skills, persona, and MCP config.
+- The parent invokes via `bajaclaw delegate <subagent> "<task>"`.
+- Permission isolation is physical: the parent doesn't have the sub-agent's
+  tools, so it literally cannot do the sub-agent's work.
+
+## Procedure
+
+### 1. Design the scope
+Ask the user:
+- **Name**: short, one-word (`mail`, `finance`, `research`, `scraper`)
+- **What it does**: one-sentence purpose
+- **What tools it needs**: e.g. `Read, Write, Bash` plus a specific MCP server
+- **What tools it must NOT have**: things the parent shouldn't do either,
+  or things scoped only to this agent
+
+### 2. Create
+```
+bajaclaw subagent create <name> \
+  --parent <main-profile> \
+  --template custom \
+  --allowed-tools Read,Write,Bash \
+  --disallowed-tools Edit \
+  --description "<one-line purpose>"
+```
+
+Any option can be omitted. Defaults: template=custom, tools=inherited.
+
+### 3. Tighten MCP servers on the sub-agent (optional)
+If the sub-agent owns a domain — say, email — you want its MCP config to
+include the relevant server. Edit:
+`~/.bajaclaw/profiles/<name>/mcp-config.json`
+
+or port a specific server from the desktop config:
+```
+bajaclaw mcp port --names email-mcp
+# then move it to the sub-agent's profile MCP file
+```
+
+Conversely: REMOVE the email MCP from the parent's MCP config so the
+parent can't use it directly. That's what makes the isolation real.
+
+### 4. Give the sub-agent its persona
+```
+bajaclaw persona --edit --profile <name>
+```
+
+Or edit `~/.bajaclaw/profiles/<name>/SOUL.md` directly. A tight persona
+that describes exactly what the sub-agent does keeps its cycles on-task.
+
+### 5. Confirm registration
+```
+bajaclaw subagent list <main-profile>
+```
+
+The new sub-agent should appear under the parent.
+
+### 6. Test the delegation path
+```
+bajaclaw delegate <name> "say hi"
+```
+
+Should print a short response from the sub-agent.
+
+From the main agent's perspective, the pattern becomes:
+```
+# In a main-agent cycle, when the task touches email:
+# The agent runs this via Bash:
+bajaclaw delegate mail "check inbox for anything from Alice"
+```
+
+## Pitfalls
+- Sub-agents don't share memory with the parent. If you want a shared
+  knowledge base, use `memorySync: true` plus careful routing.
+- Don't over-fragment. One sub-agent per clear responsibility is plenty.
+  A tree of 10 sub-agents with overlapping scopes is worse than none.
+- The parent must know to delegate. Add the `delegate-to-subagent`
+  built-in skill so its matcher picks up on the relevant user requests.
+
+## Verification
+- `bajaclaw subagent list` shows the parent → sub-agent tree.
+- `~/.bajaclaw/profiles/<parent>/config.json` lists the sub-agent in
+  `subAgents`.
+- `~/.bajaclaw/profiles/<sub>/config.json` has `parent: "<parent>"`.
+- A delegation test returns a non-empty response from the sub-agent.

package/skills/setup-telegram/SKILL.md

@@ -0,0 +1,54 @@
+---
+name: setup-telegram
+description: Walk the user through adding a Telegram bot adapter to BajaClaw
+version: 0.1.0
+tools: [Bash, Read, Write, Edit]
+triggers: ["setup telegram", "help me with telegram", "connect telegram", "telegram bot", "add telegram", "telegram setup"]
+effort: medium
+---
+
+## When to use
+The user asks you to connect Telegram, add a Telegram bot, forward DMs
+through a bot, or receive/send BajaClaw tasks over Telegram.
+
+## Quick reference
+- Adapter: `src/channels/gateway.ts`, uses `node-telegram-bot-api` (optional dep)
+- Token source: Telegram's `@BotFather` — `/newbot` → `/token`
+- Allowlist: the user's numeric Telegram user id (from `@userinfobot`)
+- Stored in: `~/.bajaclaw/profiles/<profile>/config.json` → `channels[]`
+
+## Procedure
+1. Check for an existing token. Ask: "do you already have a Telegram bot token?"
+   - If no: tell them to open Telegram, message `@BotFather`, send `/newbot`,
+     pick a name and username. BotFather replies with a token shaped like
+     `<digits>:<alphanumeric>`. Copy it.
+2. Ask for their numeric Telegram user id. They can get it by messaging
+   `@userinfobot`. Without this the allowlist is empty and no messages get
+   through.
+3. Run: `bajaclaw channel add <profile> telegram --token <TOKEN>` (replace
+   `<profile>` with their active profile — default is `default`).
+4. Edit `~/.bajaclaw/profiles/<profile>/config.json`. Inside
+   `channels[0].allowlist`, put `[<NUMERIC_USER_ID>]`. Example:
+   `"allowlist": [123456789]`
+5. Start the gateway: `bajaclaw daemon start <profile>` (or a dedicated
+   gateway subprocess if you're keeping the daemon off).
+6. Ask the user to message their bot. Expect a new task to appear in
+   `bajaclaw status <profile>` and a `gateway.telegram.msg` entry in the
+   logs.
+
+## Pitfalls
+- `npm install node-telegram-bot-api` in the BajaClaw install dir if the
+  dependency is missing (it's an `optionalDependencies` entry).
+- Empty allowlist → zero messages accepted. Confirm the id is present.
+- Bot cannot initiate a conversation — the user must message the bot first
+  OR the bot must be added to a group.
+- Never paste the token into a public log or chat. If it leaks, revoke via
+  BotFather `/revoke`.
+- `bajaclaw uninstall` removes all channel configs as part of profile
+  teardown.
+
+## Verification
+- `bajaclaw channel list <profile>` shows the telegram entry.
+- `bajaclaw daemon logs <profile> --lines 50` includes `gateway.telegram.msg`
+  after a test message.
+- `bajaclaw status <profile>` shows the pending-tasks counter rising.

package/skills/setup-uninstall/SKILL.md

@@ -0,0 +1,55 @@
+---
+name: setup-uninstall
+description: Safely tear down BajaClaw — integrations, schedulers, data, or all three
+version: 0.1.0
+tools: [Bash, Read]
+triggers: ["uninstall", "remove bajaclaw", "delete bajaclaw", "clean up bajaclaw", "tear down"]
+effort: low
+---
+
+## When to use
+User wants BajaClaw off their machine — completely, or just the
+integrations (keeping data), or just a specific profile.
+
+## Quick reference
+- Dry-run default: `bajaclaw uninstall` prints the plan, changes nothing.
+- `--yes` applies. `--keep-data` preserves `~/.bajaclaw/`.
+- Per-profile removal: `bajaclaw profile delete <name> --yes` is the
+  scoped alternative.
+
+## Procedure
+
+### Full teardown
+1. Show the plan: `bajaclaw uninstall`
+2. Review — every item listed will be removed.
+3. Apply: `bajaclaw uninstall --yes`
+4. Remove the binary itself: `npm uninstall -g bajaclaw`
+
+### Partial (keep data, remove integrations only)
+1. `bajaclaw uninstall --yes --keep-data`
+2. Removes OS scheduler entries, agent descriptors, MCP registration,
+   memory sync files. Preserves `~/.bajaclaw/` — profiles, DB, skills, logs.
+
+### Remove one profile
+1. `bajaclaw profile delete <name> --yes`
+2. Does NOT remove the global integrations (MCP registration, user-scope
+   skills) — use full uninstall for that.
+
+## What gets removed (full uninstall)
+- Running daemons (SIGTERM)
+- OS scheduler entries for every profile (launchd plist / systemd unit /
+  crontab line / schtasks)
+- `~/.claude/agents/<profile>/` for every profile
+- Memory sync files at `~/.claude/memory/bajaclaw-*.md`
+- BajaClaw's MCP entry in the desktop MCP config
+- `~/.bajaclaw/` (unless `--keep-data`)
+
+## Pitfalls
+- Uninstall is irreversible without `--keep-data`. Back up
+  `~/.bajaclaw/profiles/<name>/bajaclaw.db` first if you want history.
+- The npm uninstall step is a separate command — `bajaclaw uninstall`
+  cannot remove its own binary.
+
+## Verification
+- `ls ~/.bajaclaw/` → "no such file or directory"
+- `bajaclaw` → "command not found" after the npm uninstall step

package/skills/web-research/SKILL.md

@@ -0,0 +1,20 @@
+---
+name: web-research
+description: Search and synthesize across sources with inline citations
+version: 0.1.0
+tools: [WebSearch, WebFetch]
+triggers: ["research", "look up", "find information about"]
+effort: high
+---
+
+## Instructions
+
+1. Search with 2-3 different queries to avoid single-source bias.
+2. Fetch the top candidates. Skim before reading — if the source looks unreliable
+   (content farm, AI-generated SEO, unsigned claim), drop it.
+3. Synthesize with inline citations: every claim carries a URL.
+4. When sources conflict, show the disagreement — don't resolve it silently.
+5. Distinguish "the source says X" from "the evidence suggests X".
+
+Final output: one-paragraph summary, bulleted findings with citations, open
+questions. Do not paste large quotes. Paraphrase and cite.

package/templates/code/AGENT.md

@@ -0,0 +1,15 @@
+# Code Baja — Operating Guide (orchestrator)
+
+You are an orchestrator. You plan and delegate. **You never write code directly.**
+
+## Rules
+- For every coding task, produce a plan first: files to touch, risks, test strategy.
+- Once the plan is clear, invoke `delegateCoding` with the scoped task.
+- You receive the sub-agent's result and summarize it back.
+- Disallowed tools: Write, Edit, Bash. You can Read, Grep, Glob to inspect.
+- If the sub-agent's output looks wrong, ask for a revision — don't try to fix it yourself.
+
+## Why
+Keeping orchestration separate from execution makes cycles reviewable: you can
+see the plan before code exists, and the sub-agent session's transcript is a
+clean unit of work.

package/templates/code/HEARTBEAT.md

@@ -0,0 +1,11 @@
+# Heartbeat Schedule
+
+Edit this file with one schedule per line in the format:
+`<cron> | <task>`
+
+Examples:
+- `*/15 * * * * | Check pending tasks.`
+- `0 9 * * *    | Run daily briefing.`
+- `0 17 * * 5   | Weekly summary and plan for next week.`
+
+On first daemon boot, BajaClaw parses this file and stores entries in the DB.

package/templates/code/SOUL.md

@@ -0,0 +1,10 @@
+# {{AGENT_NAME}} — Identity
+
+You are {{AGENT_NAME}}. Edit this file to define who this agent is — its domain,
+voice, priorities, and guardrails. This section is prepended to every prompt.
+
+Describe:
+- Primary purpose (one sentence)
+- What you should and shouldn't do
+- Tone and voice
+- Who you report to

package/templates/code/bajaclaw.config.json

@@ -0,0 +1,5 @@
+{
+  "name": "{{AGENT_NAME}}",
+  "template": "{{TEMPLATE}}",
+  "notes": "Edit this file to override per-agent settings. Top-level config lives in ~/.bajaclaw/profiles/{{AGENT_NAME}}/config.json"
+}

package/templates/code/scripts/launch.bat

@@ -0,0 +1,4 @@
+@echo off
+setlocal
+cd /d "%~dp0\.."
+bajaclaw start "{{AGENT_NAME}}" %*

package/templates/code/scripts/launch.sh

@@ -0,0 +1,4 @@
+#!/usr/bin/env bash
+set -euo pipefail
+cd "$(dirname "$0")/.."
+bajaclaw start "{{AGENT_NAME}}" "$@"

package/templates/custom/AGENT.md

@@ -0,0 +1,14 @@
+# {{AGENT_NAME}} — Operating Guide
+
+You are **{{AGENT_NAME}}**, a BajaClaw agent.
+
+## How you run
+- You are invoked by the `bajaclaw` CLI as `claude -p` subprocesses.
+- Each invocation is one cycle. You do not persist memory across invocations on your own; BajaClaw injects relevant memories for you.
+- Your system prompt is assembled from: SOUL.md (identity) + this file + matched skills + recalled memories + the current task.
+
+## Rules
+- Read carefully before acting. If you need a tool you do not have, say so and stop.
+- No placeholder data. If you don't know, say so.
+- Keep final responses terse. You are a working agent, not a chatbot.
+- If you produce durable facts, they will be extracted post-cycle automatically. Include them in your response as plain sentences.

package/templates/custom/HEARTBEAT.md

@@ -0,0 +1,11 @@
+# Heartbeat Schedule
+
+Edit this file with one schedule per line in the format:
+`<cron> | <task>`
+
+Examples:
+- `*/15 * * * * | Check pending tasks.`
+- `0 9 * * *    | Run daily briefing.`
+- `0 17 * * 5   | Weekly summary and plan for next week.`
+
+On first daemon boot, BajaClaw parses this file and stores entries in the DB.

package/templates/custom/SOUL.md

@@ -0,0 +1,10 @@
+# {{AGENT_NAME}} — Identity
+
+You are {{AGENT_NAME}}. Edit this file to define who this agent is — its domain,
+voice, priorities, and guardrails. This section is prepended to every prompt.
+
+Describe:
+- Primary purpose (one sentence)
+- What you should and shouldn't do
+- Tone and voice
+- Who you report to

package/templates/custom/bajaclaw.config.json

@@ -0,0 +1,5 @@
+{
+  "name": "{{AGENT_NAME}}",
+  "template": "{{TEMPLATE}}",
+  "notes": "Edit this file to override per-agent settings. Top-level config lives in ~/.bajaclaw/profiles/{{AGENT_NAME}}/config.json"
+}

package/templates/custom/scripts/launch.bat

@@ -0,0 +1,4 @@
+@echo off
+setlocal
+cd /d "%~dp0\.."
+bajaclaw start "{{AGENT_NAME}}" %*

package/templates/custom/scripts/launch.sh

@@ -0,0 +1,4 @@
+#!/usr/bin/env bash
+set -euo pipefail
+cd "$(dirname "$0")/.."
+bajaclaw start "{{AGENT_NAME}}" "$@"

package/templates/outreach/AGENT.md

@@ -0,0 +1,11 @@
+# Outreach Baja — Operating Guide
+
+You are an outreach agent. You draft, you never send without approval.
+
+## Rules
+- Always produce drafts in plain text. Subject + body separated clearly.
+- Personalize. Generic outreach is worse than no outreach.
+- Keep messages under 150 words unless the recipient has asked for detail.
+- Include a clear ask. One ask per message.
+- Follow-up cadence: day 0, day 4, day 11. Three-strike rule — stop after three ignored messages.
+- Never invent relationships, quotes, or mutual contacts.

package/templates/outreach/HEARTBEAT.md

@@ -0,0 +1,11 @@
+# Heartbeat Schedule
+
+Edit this file with one schedule per line in the format:
+`<cron> | <task>`
+
+Examples:
+- `*/15 * * * * | Check pending tasks.`
+- `0 9 * * *    | Run daily briefing.`
+- `0 17 * * 5   | Weekly summary and plan for next week.`
+
+On first daemon boot, BajaClaw parses this file and stores entries in the DB.

package/templates/outreach/SOUL.md

@@ -0,0 +1,10 @@
+# {{AGENT_NAME}} — Identity
+
+You are {{AGENT_NAME}}. Edit this file to define who this agent is — its domain,
+voice, priorities, and guardrails. This section is prepended to every prompt.
+
+Describe:
+- Primary purpose (one sentence)
+- What you should and shouldn't do
+- Tone and voice
+- Who you report to

package/templates/outreach/bajaclaw.config.json

@@ -0,0 +1,5 @@
+{
+  "name": "{{AGENT_NAME}}",
+  "template": "{{TEMPLATE}}",
+  "notes": "Edit this file to override per-agent settings. Top-level config lives in ~/.bajaclaw/profiles/{{AGENT_NAME}}/config.json"
+}

package/templates/outreach/scripts/launch.bat

@@ -0,0 +1,4 @@
+@echo off
+setlocal
+cd /d "%~dp0\.."
+bajaclaw start "{{AGENT_NAME}}" %*

package/templates/outreach/scripts/launch.sh

@@ -0,0 +1,4 @@
+#!/usr/bin/env bash
+set -euo pipefail
+cd "$(dirname "$0")/.."
+bajaclaw start "{{AGENT_NAME}}" "$@"

package/templates/research/AGENT.md

@@ -0,0 +1,27 @@
+# Research Baja — Operating Guide
+
+You are a research agent. You **gather, synthesize, and act on findings** —
+you are not a read-only observer. Research leads to decisions, decisions lead
+to artifacts. Produce the artifacts.
+
+## Rules
+- Cite sources with URL. No citation → no claim.
+- Prefer primary sources over secondary. If a claim rests on one source, say so.
+- When sources disagree, surface the disagreement; do not pick a side without reason.
+- Distinguish "I found X" from "I infer Y from X".
+- When the user asks for an artifact (report, summary, draft, plan, script,
+  code change), produce it. Don't just describe what you'd do.
+
+## Tools
+You have full tool access: `Read`, `Write`, `Edit`, `Bash`, `Grep`, `Glob`,
+`WebSearch`, `WebFetch`, plus whatever MCP tools are configured. Use them as
+needed to complete the work.
+
+## Output format
+When the task is a pure question, answer it:
+1. One-paragraph executive summary.
+2. Key findings with inline citations.
+3. Open questions and what you'd need to answer them.
+
+When the task is a production task (write this, build that, fix this):
+produce the artifact as the main output, then note what you used to make it.

package/templates/research/HEARTBEAT.md

@@ -0,0 +1,11 @@
+# Heartbeat Schedule
+
+Edit this file with one schedule per line in the format:
+`<cron> | <task>`
+
+Examples:
+- `*/15 * * * * | Check pending tasks.`
+- `0 9 * * *    | Run daily briefing.`
+- `0 17 * * 5   | Weekly summary and plan for next week.`
+
+On first daemon boot, BajaClaw parses this file and stores entries in the DB.

package/templates/research/SOUL.md

@@ -0,0 +1,10 @@
+# {{AGENT_NAME}} — Identity
+
+You are {{AGENT_NAME}}. Edit this file to define who this agent is — its domain,
+voice, priorities, and guardrails. This section is prepended to every prompt.
+
+Describe:
+- Primary purpose (one sentence)
+- What you should and shouldn't do
+- Tone and voice
+- Who you report to

package/templates/research/bajaclaw.config.json

@@ -0,0 +1,5 @@
+{
+  "name": "{{AGENT_NAME}}",
+  "template": "{{TEMPLATE}}",
+  "notes": "Edit this file to override per-agent settings. Top-level config lives in ~/.bajaclaw/profiles/{{AGENT_NAME}}/config.json"
+}

package/templates/research/scripts/launch.bat

@@ -0,0 +1,4 @@
+@echo off
+setlocal
+cd /d "%~dp0\.."
+bajaclaw start "{{AGENT_NAME}}" %*

package/templates/research/scripts/launch.sh

@@ -0,0 +1,4 @@
+#!/usr/bin/env bash
+set -euo pipefail
+cd "$(dirname "$0")/.."
+bajaclaw start "{{AGENT_NAME}}" "$@"

package/templates/social/AGENT.md

@@ -0,0 +1,10 @@
+# Social Baja — Operating Guide
+
+You draft and schedule social content. You do not post without approval.
+
+## Rules
+- Match the platform's native voice — don't cross-post verbatim.
+- One idea per post. Thread when the idea is bigger than one post.
+- Never use engagement-bait hooks.
+- No claims without support. No fake statistics.
+- Save drafts to the tasks queue; flag the best for user review.

package/templates/social/HEARTBEAT.md

@@ -0,0 +1,11 @@
+# Heartbeat Schedule
+
+Edit this file with one schedule per line in the format:
+`<cron> | <task>`
+
+Examples:
+- `*/15 * * * * | Check pending tasks.`
+- `0 9 * * *    | Run daily briefing.`
+- `0 17 * * 5   | Weekly summary and plan for next week.`
+
+On first daemon boot, BajaClaw parses this file and stores entries in the DB.

package/templates/social/SOUL.md

@@ -0,0 +1,10 @@
+# {{AGENT_NAME}} — Identity
+
+You are {{AGENT_NAME}}. Edit this file to define who this agent is — its domain,
+voice, priorities, and guardrails. This section is prepended to every prompt.
+
+Describe:
+- Primary purpose (one sentence)
+- What you should and shouldn't do
+- Tone and voice
+- Who you report to