bajaclaw 0.14.3 → 0.14.5

package/README.md

@@ -7,17 +7,17 @@
  ██╔══██╗██╔══██║██   ██║██╔══██║    ██║     ██║     ██╔══██║██║███╗██║
  ██████╔╝██║  ██║╚█████╔╝██║  ██║    ╚██████╗███████╗██║  ██║╚███╔███╔╝
  ╚═════╝ ╚═╝  ╚═╝ ╚════╝ ╚═╝  ╚═╝     ╚═════╝╚══════╝╚═╝  ╚═╝ ╚══╝╚══╝
-          autonomous agents on your terms  ·  MIT  ·  v0.14.3
+          autonomous agents on your terms  ·  MIT  ·  v0.14.5
 ```
 
 **BajaClaw is a long-running agent runtime for the `claude` CLI.** It turns
 the one-shot `claude -p` command into an always-on, scheduled, memory-backed,
-skill-matching autonomous agent — with a local dashboard, multiple profiles,
+skill-matching autonomous agent - with a local dashboard, multiple profiles,
 OS-native scheduling, and first-class MCP integration.
 
 You install it once. It sets itself up. You run `bajaclaw start`. It goes.
 
-> The name is a tribute to Baja Blast — the author's favorite soda. The
+> The name is a tribute to Baja Blast - the author's favorite soda. The
 > dashboard ships in that same tropical-lime teal on purpose.
 
 ---
@@ -36,7 +36,7 @@ That's it. The post-install runs `bajaclaw setup` automatically, which:
 - Runs the health check and tells you if anything's off
 
 **Requirements**: Node 20+ and the `claude` CLI backend on your `PATH`.
-BajaClaw drives the backend as a subprocess — whatever login/subscription
+BajaClaw drives the backend as a subprocess - whatever login/subscription
 that CLI uses is what BajaClaw uses. BajaClaw itself never sees credentials.
 
 First run, end-to-end:
@@ -82,7 +82,7 @@ A BajaClaw **cycle** is one pass of the 13-step loop in
 
 You can run a cycle manually (`bajaclaw start`), schedule it (`bajaclaw
 daemon install`), or trigger it externally (`bajaclaw trigger <event>`).
-Cycles are idempotent — safe to re-run.
+Cycles are idempotent - safe to re-run.
 
 ---
 
@@ -101,9 +101,9 @@ You can invoke it from a Claude Code session the same way you invoke any
 other agent. The BajaClaw profile is the durable side; the Claude Code
 descriptor is the handle.
 
-### Claude Code skills — compatible format, **isolated scope**
+### Claude Code skills - compatible format, **isolated scope**
 
-A BajaClaw skill is a `SKILL.md` file with YAML frontmatter — byte-for-byte
+A BajaClaw skill is a `SKILL.md` file with YAML frontmatter - byte-for-byte
 the same shape Claude Code uses. But BajaClaw reads only BajaClaw-owned
 directories:
 
@@ -114,7 +114,7 @@ directories:
 | 3 | `~/.bajaclaw/skills/` |
 | 4 | `<repo>/skills/` (built-ins) |
 
-`~/.claude/skills/` is **not** read automatically — that keeps the two
+`~/.claude/skills/` is **not** read automatically - that keeps the two
 agents from stepping on each other's skill libraries.
 
 **Porting skills from Claude Code is a one-liner:**
@@ -137,7 +137,7 @@ Skills`. See [`docs/skills.md`](docs/skills.md).
 ### Self-knowledge (built-in guides)
 
 BajaClaw knows how to configure itself. Ship 13 built-in skills describe
-the procedure for every integration — ask your agent in plain language:
+the procedure for every integration - ask your agent in plain language:
 
 > "Help me set up Telegram."
 > "Walk me through connecting Discord."
@@ -171,7 +171,7 @@ Procedure / Pitfalls / Verification sections into
 `~/.bajaclaw/skills/auto/<name>/` for review.
 
 This is BajaClaw's take on the "create a skill after a complex task"
-behavior from agents like Hermes — capture procedures the first time you
+behavior from agents like Hermes - capture procedures the first time you
 solve them so repeats are faster.
 
 Candidates live in `auto/` until you promote them:
@@ -187,7 +187,7 @@ Tune the trigger in the profile's `config.json`:
 { "autoSkill": { "enabled": true, "minToolUses": 5, "maxPerDay": 10 } }
 ```
 
-### MCP — isolated by default
+### MCP - isolated by default
 
 BajaClaw uses its own MCP config. The desktop CLI's `mcpServers` is **not**
 inherited by default. Merge order per cycle (highest wins):
@@ -195,7 +195,7 @@ inherited by default. Merge order per cycle (highest wins):
 1. `<profile>/agent-mcp-config.json`
 2. `<profile>/mcp-config.json`
 3. `~/.bajaclaw/mcp-config.json` (user-global BajaClaw)
-4. Desktop CLI config — **only if `mergeDesktopMcp: true`** in the profile
+4. Desktop CLI config - **only if `mergeDesktopMcp: true`** in the profile
 
 Port servers from Claude Code on demand:
 
@@ -212,7 +212,7 @@ desktop MCP list on every cycle (pre-0.4 behavior).
 BajaClaw's own MCP entry (`bajaclaw`) is skipped automatically during port
 to avoid self-references.
 
-### MCP — expose
+### MCP - expose
 
 BajaClaw is itself an MCP server. `bajaclaw setup` auto-registers it so your
 desktop MCP client (Claude Desktop and anything else that reads that config)
@@ -220,10 +220,10 @@ can query BajaClaw's state directly.
 
 **Resources:**
 
-- `bajaclaw://profiles` — list of profiles
+- `bajaclaw://profiles` - list of profiles
 - `bajaclaw://profile/<n>/agents`
-- `bajaclaw://profile/<n>/memories` — FTS5-searchable
-- `bajaclaw://profile/<n>/cycles` — recent cycle history
+- `bajaclaw://profile/<n>/memories` - FTS5-searchable
+- `bajaclaw://profile/<n>/cycles` - recent cycle history
 - `bajaclaw://profile/<n>/schedules`
 
 **Tools:**
@@ -234,7 +234,7 @@ can query BajaClaw's state directly.
 - `bajaclaw_skill_list({ profile })`
 
 Which means: from any MCP client, you can ask "what does BajaClaw remember
-about X?" or "queue a task for the default agent" — without leaving your
+about X?" or "queue a task for the default agent" - without leaving your
 current session.
 
 Run it manually with `bajaclaw mcp serve --stdio` or as HTTP SSE with
@@ -249,7 +249,7 @@ Set `memorySync: true` in the profile config and BajaClaw will:
 - Write a digest to `~/.claude/memory/bajaclaw-<profile>.md` after each
   cycle, so Claude Code sessions can see what BajaClaw has been learning
 
-Disabled by default — memory sharing is deliberate, not automatic. See
+Disabled by default - memory sharing is deliberate, not automatic. See
 [`docs/memory.md`](docs/memory.md).
 
 ### Sub-agent delegation
@@ -257,7 +257,7 @@ Disabled by default — memory sharing is deliberate, not automatic. See
 For heavy coding work, an agent using the `code` template plans and then
 delegates to a dedicated Claude Code sub-session via `delegateCoding` in
 [`src/delegation.ts`](src/delegation.ts). The orchestrator never writes code
-itself — that keeps its transcript reviewable before any code exists. See
+itself - that keeps its transcript reviewable before any code exists. See
 [`docs/integration.md`](docs/integration.md).
 
 ---
@@ -267,12 +267,12 @@ itself — that keeps its transcript reviewable before any code exists. See
 ```
 bajaclaw start                 # runs a cycle against the default profile
 bajaclaw start --dry-run       # shows the assembled prompt + exact argv
-bajaclaw dashboard             # http://localhost:7337 — live cycle feed, memories
+bajaclaw dashboard             # http://localhost:7337 - live cycle feed, memories
 bajaclaw daemon install        # schedule a 15-minute heartbeat via your OS
 bajaclaw daemon start          # supervisor loop with exponential backoff
 ```
 
-The default profile has **full tool access** — Read, Write, Edit, Bash,
+The default profile has **full tool access** - Read, Write, Edit, Bash,
 Grep, Glob, WebSearch, WebFetch, plus every MCP tool you've configured. It's
 a real autonomous agent, not a sandboxed assistant.
 
@@ -313,7 +313,7 @@ BAJACLAW_PROFILE=triage bajaclaw daemon start
 
 | template | shape |
 |---|---|
-| `custom`   | blank slate, full tools — the default |
+| `custom`   | blank slate, full tools - the default |
 | `research` | research + synthesis + artifacts; full tools |
 | `outreach` | email prospecting + drafting |
 | `support`  | inbox triage + reply drafts |
@@ -333,7 +333,7 @@ classifies the task and routes it to the cheapest capable model:
 | **Sonnet** | normal work, answers, summaries | 5 memories, 2 skills, 8 turns |
 | **Opus**   | planning, coding, refactoring, deep research, reflection | 7 memories, 3 skills, 14 turns |
 
-The classifier is a heuristic — zero extra backend calls for routing.
+The classifier is a heuristic - zero extra backend calls for routing.
 Post-cycle memory extract + auto-skill synthesis are **skipped**
 entirely for Haiku cycles. This keeps cheap tasks cheap.
 
@@ -353,8 +353,8 @@ bajaclaw serve --api-key $(openssl rand -hex 32) # with auth
 bajaclaw serve --host 0.0.0.0 --api-key <key>    # bind all (auth required)
 ```
 
-Anything that speaks the OpenAI chat API — Cursor, Open WebUI, LibreChat,
-`openai` SDKs, curl, LangChain, LlamaIndex — can drive BajaClaw as an LLM.
+Anything that speaks the OpenAI chat API - Cursor, Open WebUI, LibreChat,
+`openai` SDKs, curl, LangChain, LlamaIndex - can drive BajaClaw as an LLM.
 The request's `model` field is a profile name; each request is one full
 cycle (memory recall, skill matching, MCP inheritance, backend call,
 post-cycle extract).
@@ -415,14 +415,14 @@ the notice with `BAJACLAW_NO_UPDATE_NOTICE=1`.
 ```
 bajaclaw setup                 # idempotent bootstrap; safe to re-run
 bajaclaw setup --profile foo   # use a different default profile name
-bajaclaw uninstall             # dry-run — shows what would be removed
+bajaclaw uninstall             # dry-run - shows what would be removed
 bajaclaw uninstall --yes       # actually tear everything down
 bajaclaw uninstall --yes --keep-data  # remove integrations, keep ~/.bajaclaw/
 ```
 
 `setup` is the re-run button. If the MCP registration gets knocked out of
 the desktop config, or the agent descriptor is missing, or you moved your
-home directory — `bajaclaw setup` fixes it all without touching existing
+home directory - `bajaclaw setup` fixes it all without touching existing
 data.
 
 `uninstall` tears down everything BajaClaw has created:
@@ -435,7 +435,7 @@ data.
 - Removes `~/.claude/memory/bajaclaw-*.md` sync files
 - Removes `~/.bajaclaw/` entirely (unless `--keep-data`)
 
-It does **not** `npm uninstall` itself — that's one command you still run by
+It does **not** `npm uninstall` itself - that's one command you still run by
 hand, printed at the end of the teardown.
 
 ---
@@ -483,7 +483,7 @@ Memories`. After the cycle finishes, a 1-turn Haiku call reads the
 Those facts become FTS-indexed rows with `source=cycle` and
 `source_cycle_id=<id>`. Next cycle, they're eligible for recall again.
 
-Kinds are a soft taxonomy — BajaClaw doesn't enforce them: `fact`,
+Kinds are a soft taxonomy - BajaClaw doesn't enforce them: `fact`,
 `decision`, `preference`, `todo`, `reference`, `claude-code`, `imported`.
 
 Full detail: [`docs/memory.md`](docs/memory.md).
@@ -493,7 +493,7 @@ Full detail: [`docs/memory.md`](docs/memory.md).
 ## Channels (optional)
 
 BajaClaw ships optional adapters for **Telegram** and **Discord** bots.
-They're `optionalDependencies` — not installed unless you use them.
+They're `optionalDependencies` - not installed unless you use them.
 
 ```
 bajaclaw channel add default telegram --token <BOT_TOKEN>
@@ -519,7 +519,7 @@ bajaclaw dashboard
 Single HTML file served at `http://localhost:7337` (port in `config.json`).
 Dark theme, vanilla JS, Tailwind CDN. Live cycle feed, FTS-searchable
 memory browser, schedule editor, inbox/tasks list. Reads directly from the
-SQLite DB — no extra service.
+SQLite DB - no extra service.
 
 ---
 
@@ -627,20 +627,20 @@ Deeper in [`docs/architecture.md`](docs/architecture.md).
 
 ## Docs
 
-- [`architecture.md`](docs/architecture.md) — module map, cycle, on-disk layout
-- [`integration.md`](docs/integration.md) — Claude Code + MCP seams in detail
-- [`commands.md`](docs/commands.md) — full command reference
-- [`agents.md`](docs/agents.md) — profiles, templates, AGENT.md / SOUL.md / HEARTBEAT.md
-- [`skills.md`](docs/skills.md) — scoping, matching, self-generated skills
-- [`memory.md`](docs/memory.md) — FTS5 recall + extract, cross-tool sync
-- [`heartbeat.md`](docs/heartbeat.md) — scheduling + supervisor
+- [`architecture.md`](docs/architecture.md) - module map, cycle, on-disk layout
+- [`integration.md`](docs/integration.md) - Claude Code + MCP seams in detail
+- [`commands.md`](docs/commands.md) - full command reference
+- [`agents.md`](docs/agents.md) - profiles, templates, AGENT.md / SOUL.md / HEARTBEAT.md
+- [`skills.md`](docs/skills.md) - scoping, matching, self-generated skills
+- [`memory.md`](docs/memory.md) - FTS5 recall + extract, cross-tool sync
+- [`heartbeat.md`](docs/heartbeat.md) - scheduling + supervisor
 - [`channels.md`](docs/channels.md)
-- [`api.md`](docs/api.md) — OpenAI-compatible HTTP endpoint — Telegram + Discord
-- [`security.md`](docs/security.md) — threat model + mitigations
-- [`fair-use.md`](docs/fair-use.md) — how BajaClaw stays a thin wrapper
-- [`troubleshooting.md`](docs/troubleshooting.md) — common fixes
-- [`faq.md`](docs/faq.md) — frequently asked
-- [`contributing.md`](docs/contributing.md) — dev setup, style, release
+- [`api.md`](docs/api.md) - OpenAI-compatible HTTP endpoint - Telegram + Discord
+- [`security.md`](docs/security.md) - threat model + mitigations
+- [`fair-use.md`](docs/fair-use.md) - how BajaClaw stays a thin wrapper
+- [`troubleshooting.md`](docs/troubleshooting.md) - common fixes
+- [`faq.md`](docs/faq.md) - frequently asked
+- [`contributing.md`](docs/contributing.md) - dev setup, style, release
 
 ---
 

package/bin/bajaclaw.js

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-// Thin launcher — resolves tsx and invokes src/cli.ts, or dist/cli.js if built.
+// Thin launcher - resolves tsx and invokes src/cli.ts, or dist/cli.js if built.
 import { spawnSync } from "node:child_process";
 import { existsSync } from "node:fs";
 import { dirname, join } from "node:path";

package/bin/create-bajaclaw.js

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-// npx create-bajaclaw [name] — bootstrap BajaClaw.
+// npx create-bajaclaw [name] - bootstrap BajaClaw.
 // With no args: runs `bajaclaw setup` to create the default profile.
 // With a name: runs `bajaclaw init <name>` to scaffold that specific profile.
 import { spawnSync } from "node:child_process";

package/dist/agent.js

@@ -163,7 +163,7 @@ async function runCycleInner(input) {
         }
         recordSuccess(db);
         // Store up to 8k chars of the response. Needed for conversational
-        // history on channel-sourced cycles — the old 300-char cap made
+        // history on channel-sourced cycles - the old 300-char cap made
         // every prior turn look like a stub. 8k ≈ 2k tokens; beyond that
         // we rely on memory extraction to preserve context.
         db.prepare("UPDATE cycles SET finished_at=?, status=?, response_preview=?, cost_usd=?, input_tokens=?, output_tokens=?, turns=? WHERE id=?").run(finished, "ok", result.text.slice(0, 8000), result.costUsd ?? null, result.inputTokens ?? null, result.outputTokens ?? null, result.turns ?? null, cycleId);
@@ -171,7 +171,7 @@ async function runCycleInner(input) {
             db.prepare("UPDATE tasks SET status='done', cycle_id=? WHERE id=?").run(cycleId, popped.id);
         }
         // Post-cycle memory extraction + auto-skill synthesis. Both are extra
-        // backend calls — skip them on cheap (Haiku) cycles and on trivially
+        // backend calls - skip them on cheap (Haiku) cycles and on trivially
         // short responses to keep token usage tight.
         const shouldDoPostWork = !result.dryRun && result.text.length >= 120 && picked.tier !== "haiku";
         if (shouldDoPostWork) {

package/dist/api/server.js

@@ -2,10 +2,10 @@
 //
 // Endpoints:
 //   GET  /health
-//   GET  /v1/models                    — lists exposed profiles as models
-//   POST /v1/chat/completions          — OpenAI chat (stream + non-stream)
-//   POST /v1/bajaclaw/cycle            — native: { profile, task } -> full CycleOutput
-//   POST /v1/bajaclaw/tasks            — native: enqueue without waiting
+//   GET  /v1/models                    - lists exposed profiles as models
+//   POST /v1/chat/completions          - OpenAI chat (stream + non-stream)
+//   POST /v1/bajaclaw/cycle            - native: { profile, task } -> full CycleOutput
+//   POST /v1/bajaclaw/tasks            - native: enqueue without waiting
 //
 // Auth: if config.api.apiKey is set, require `Authorization: Bearer <key>`.
 // Bind: defaults to 127.0.0.1 unless config.api.host is explicit.
@@ -139,18 +139,18 @@ function listProfilesAsModels(exposed) {
     const names = filter ? all.filter((n) => filter.has(n)) : all;
     const now = Math.floor(Date.now() / 1000);
     const out = [];
-    // Bare profile names — use each profile's configured model.
+    // Bare profile names - use each profile's configured model.
     for (const id of names) {
         out.push({ id, object: "model", created: now, owned_by: "bajaclaw" });
     }
-    // profile:model virtual entries — pick any model per request without
+    // profile:model virtual entries - pick any model per request without
     // touching profile config.
     for (const id of names) {
         for (const m of KNOWN_MODELS) {
             out.push({ id: `${id}:${m.id}`, object: "model", created: now, owned_by: "bajaclaw" });
         }
     }
-    // Bare model-id shortcuts — apply to the default profile.
+    // Bare model-id shortcuts - apply to the default profile.
     if (names.includes("default")) {
         for (const m of KNOWN_MODELS) {
             out.push({ id: m.id, object: "model", created: now, owned_by: "bajaclaw" });

package/dist/channels/gateway.d.ts

@@ -5,14 +5,14 @@ export declare function startAllGateways(profile: string): Promise<void>;
 /** Backwards-compatible entry point: starts adapters, then blocks. */
 export declare function runGateway(profile: string): Promise<void>;
 /** Send an agent reply back to whatever channel originated a task.
- *  `source` is formatted as "telegram:<id>" or "discord:<id>" — the
+ *  `source` is formatted as "telegram:<id>" or "discord:<id>" - the
  *  same string written into tasks.source by the inbound handlers.
  *  Also ends any typing indicator associated with that source. */
 export declare function replyToSource(profile: string, source: string, text: string): Promise<void>;
 /** Show the platform's "typing…" indicator for the given source. The
  *  adapter internally refreshes on the platform's cadence (Telegram
  *  indicator lasts ~5s, Discord ~10s) until `endTyping(source)` is
- *  called. Safe to call multiple times — a second call replaces the
+ *  called. Safe to call multiple times - a second call replaces the
  *  first. No-ops if the adapter isn't loaded. */
 export declare function beginTyping(profile: string, source: string): void;
 export declare function endTyping(source: string): void;

package/dist/channels/gateway.js

@@ -54,7 +54,7 @@ export async function runGateway(profile) {
     await new Promise(() => { });
 }
 /** Send an agent reply back to whatever channel originated a task.
- *  `source` is formatted as "telegram:<id>" or "discord:<id>" — the
+ *  `source` is formatted as "telegram:<id>" or "discord:<id>" - the
  *  same string written into tasks.source by the inbound handlers.
  *  Also ends any typing indicator associated with that source. */
 export async function replyToSource(profile, source, text) {
@@ -74,7 +74,7 @@ export async function replyToSource(profile, source, text) {
 /** Show the platform's "typing…" indicator for the given source. The
  *  adapter internally refreshes on the platform's cadence (Telegram
  *  indicator lasts ~5s, Discord ~10s) until `endTyping(source)` is
- *  called. Safe to call multiple times — a second call replaces the
+ *  called. Safe to call multiple times - a second call replaces the
  *  first. No-ops if the adapter isn't loaded. */
 export function beginTyping(profile, source) {
     const colon = source.indexOf(":");
@@ -94,7 +94,7 @@ export function beginTyping(profile, source) {
         const stop = a.startTyping(id);
         activeTyping.set(source, stop);
     }
-    catch { /* ignore — typing is best-effort */ }
+    catch { /* ignore - typing is best-effort */ }
 }
 export function endTyping(source) {
     const stop = activeTyping.get(source);
@@ -149,7 +149,7 @@ async function startTelegram(profile, c, log) {
         startTyping: (chatId) => {
             // Telegram's typing indicator auto-clears after 5s, so re-send
             // every 4s until stopped. `sendChatAction` errors are swallowed
-            // — they'd be noise (bot blocked, etc.) and the agent still has
+            // - they'd be noise (bot blocked, etc.) and the agent still has
             // a reply path via the eventual send.
             const send = () => {
                 bot.sendChatAction(Number(chatId), "typing").catch(() => undefined);

package/dist/claude.js

@@ -57,7 +57,7 @@ export function buildCommand(prompt, opts) {
     // --effort is the real knob for "how much runway does the agent get".
     // Levels: low < medium < high < xhigh < max. Higher = more turns
     // + more tokens + higher cost. claude's internal turn budget is
-    // tied to this level — there is no `--max-turns` flag.
+    // tied to this level - there is no `--max-turns` flag.
     if (opts.effort)
         args.push("--effort", opts.effort);
     if (opts.allowedTools?.length)
@@ -69,7 +69,7 @@ export function buildCommand(prompt, opts) {
     if (opts.systemPrompt)
         args.push("--system-prompt", opts.systemPrompt);
     // Beta flags. `context1M: true` is shorthand for adding the
-    // `context-1m-2025-08-07` beta. API-key-only — the CLI warns and
+    // `context-1m-2025-08-07` beta. API-key-only - the CLI warns and
     // falls back to 200k for subscription auth.
     const betas = [...(opts.betas ?? [])];
     if (opts.context1M && !betas.includes("context-1m-2025-08-07")) {
@@ -78,7 +78,7 @@ export function buildCommand(prompt, opts) {
     if (betas.length > 0)
         args.push("--betas", ...betas);
     // Per-cycle cost ceiling. Safer than a turn cap because agent
-    // complexity varies wildly — this caps the actual spend, not a
+    // complexity varies wildly - this caps the actual spend, not a
     // proxy for it.
     if (opts.maxBudgetUsd !== undefined) {
         args.push("--max-budget-usd", String(opts.maxBudgetUsd));
@@ -128,7 +128,7 @@ export async function runOnce(prompt, opts = {}) {
             // Explicitly close stdin. Without this, the claude CLI waits 3s
             // for piped input before proceeding (and treats the wait as a
             // warning that contaminates stdout). Closing stdin tells it
-            // "the prompt on -p is complete — don't wait for more".
+            // "the prompt on -p is complete - don't wait for more".
             stdin: "ignore",
         });
         return parseResult(r.stdout, r.stderr, r.exitCode ?? 0, start, jsonSupported, ["claude", ...cmd]);
@@ -163,12 +163,12 @@ function parseResult(stdout, stderr, exitCode, start, jsonMode, command) {
         command,
     };
     // True when the JSON result block explicitly reported success. When
-    // it's set, trust the JSON over a non-zero exit code — claude can
+    // it's set, trust the JSON over a non-zero exit code - claude can
     // exit non-zero if a child process it spawned (e.g. a long-running
     // dashboard server) leaves stdout/stderr pipes open past the
     // parent's timeout, even though the turn itself completed cleanly.
     let jsonReportedSuccess = false;
-    // Try JSON parse regardless of exit code — claude sometimes emits
+    // Try JSON parse regardless of exit code - claude sometimes emits
     // useful error detail in JSON even when exiting non-zero.
     if (jsonMode) {
         try {
@@ -251,7 +251,7 @@ function parseResult(stdout, stderr, exitCode, start, jsonMode, command) {
     }
     if (exitCode !== 0 && !base.error && !jsonReportedSuccess) {
         // Prefer stderr, then stdout first-line fallback. Never echo the
-        // entire stdout into the error field — it may be a multi-KB JSON.
+        // entire stdout into the error field - it may be a multi-KB JSON.
         const trimmedStderr = stderr.trim();
         if (trimmedStderr)
             base.error = trimmedStderr.slice(0, 400);
@@ -263,7 +263,7 @@ function parseResult(stdout, stderr, exitCode, start, jsonMode, command) {
     // If we determined an error from JSON but exit was 0, still mark not-ok.
     if (base.error && base.ok)
         base.ok = false;
-    // Conversely: JSON explicitly said success. Honor that — ignore a
+    // Conversely: JSON explicitly said success. Honor that - ignore a
     // non-zero exit caused by lingering child processes.
     if (jsonReportedSuccess) {
         base.ok = true;

package/dist/cli.js

@@ -35,7 +35,7 @@ function defaultProfile(explicit) {
     return explicit ?? process.env.BAJACLAW_PROFILE ?? DEFAULT_PROFILE_NAME;
 }
 const program = new Command();
-program.name(pkg.name).description("BajaClaw — autonomous agents on your terms").version(pkg.version);
+program.name(pkg.name).description("BajaClaw - autonomous agents on your terms").version(pkg.version);
 program
     .command("init [name]")
     .description("Scaffold a new agent profile")
@@ -53,7 +53,7 @@ program
 });
 program
     .command("chat [profile]")
-    .description("Interactive chat REPL — converse with the agent turn-by-turn")
+    .description("Interactive chat REPL - converse with the agent turn-by-turn")
     .option("--model <id>", "model or alias (auto|haiku|sonnet|opus|<full-id>)")
     .action(async (p, opts) => {
     const target = defaultProfile(p);
@@ -188,7 +188,7 @@ program.command("update").description("Check for and install a newer version")
     .option("--check", "only check; don't install")
     .option("--yes", "apply without confirmation")
     .action(async (opts) => runUpdate({ check: !!opts.check, yes: !!opts.yes }));
-// Setup — idempotent first-run bootstrap. Safe to rerun. Interactive
+// Setup - idempotent first-run bootstrap. Safe to rerun. Interactive
 // persona wizard on a TTY the first time; non-interactive otherwise.
 program.command("setup").description("Idempotent first-run bootstrap (profile, MCP register, persona wizard, health check)")
     .option("--profile <name>", "profile name (default: 'default')")
@@ -207,7 +207,7 @@ program.command("setup").description("Idempotent first-run bootstrap (profile, M
     interactive: !!opts.interactive,
     nonInteractive: !!opts.nonInteractive,
 }));
-// Compact — run memory compaction (or show what would run)
+// Compact - run memory compaction (or show what would run)
 program.command("compact [profile]")
     .description("Compact memory (summarize old entries, prune stale cycle rows)")
     .option("--dry-run", "show trigger state and policy without running")
@@ -242,7 +242,7 @@ program.command("compact [profile]")
         set: Object.keys(set).length > 0 ? set : undefined,
     });
 });
-// Persona — view or re-run the wizard
+// Persona - view or re-run the wizard
 program.command("persona [profile]")
     .description("Show or edit the agent's persona (name, tone, user name, focus)")
     .option("--edit", "re-run the interactive wizard")
@@ -270,30 +270,30 @@ subCmd.command("create <name>")
 subCmd.command("list [parent]")
     .description("List sub-agents under a parent (or the whole tree if omitted)")
     .action(async (p) => subagent.cmdList(p));
-// Delegate — orchestrators call this via Bash to hand off a task
+// Delegate - orchestrators call this via Bash to hand off a task
 program.command("delegate <subagent> <task>")
     .description("Run one cycle on a sub-agent and stream its response text to stdout")
     .option("--json", "output full CycleOutput JSON instead of just text")
     .action(async (sub, task, opts) => subagent.cmdDelegate(sub, task, { json: !!opts.json }));
-// Uninstall — full teardown.
+// Uninstall - full teardown.
 program.command("uninstall").description("Remove all BajaClaw state (profiles, scheduler, MCP, memory sync)")
     .option("--yes", "actually perform the teardown")
     .option("--keep-data", "keep ~/.bajaclaw/ data; only remove integrations")
     .action(async (opts) => runUninstall({ yes: !!opts.yes, keepData: !!opts.keepData }));
-// Model — show or set per profile
+// Model - show or set per profile
 program.command("model [value] [profile]")
     .description("Show or set the model for a profile (no value: lists known models)")
     .action(async (value, p) => runModel(value, { profile: p }));
-// Effort — show or set per profile
+// Effort - show or set per profile
 program.command("effort [value] [profile]")
     .description("Show or set the effort level (low/medium/high) for a profile")
     .action(async (value, p) => runEffort(value, { profile: p }));
-// Guide — print a self-setup walkthrough
+// Guide - print a self-setup walkthrough
 program.command("guide [topic]")
     .description("Print a self-setup walkthrough, or list available guides")
     .option("--profile <name>", "profile to use for skill lookup")
     .action(async (topic, opts) => runGuide(topic, { profile: opts.profile }));
-// Serve — OpenAI-compatible HTTP endpoint
+// Serve - OpenAI-compatible HTTP endpoint
 program.command("serve")
     .description("Serve BajaClaw over an OpenAI-compatible HTTP API")
     .option("--host <host>", "bind host (default 127.0.0.1)")
@@ -312,7 +312,7 @@ program.command("serve")
 program.command("banner").description("Print the ASCII banner").action(() => {
     printBanner(pkg.version, { force: true });
 });
-// Welcome — first-run greeting; also callable anytime
+// Welcome - first-run greeting; also callable anytime
 program.command("welcome").description("Print the welcome banner + next steps")
     .action(async () => {
     await printWelcome({ force: true });
@@ -341,7 +341,7 @@ async function printWelcome(opts = {}) {
     }
     catch { /* health check optional */ }
     console.log(chalk.bold("Start chatting:"));
-    console.log(`  ${chalk.cyan("bajaclaw chat")}                   ${chalk.dim("# interactive REPL — talk to your agent")}`);
+    console.log(`  ${chalk.cyan("bajaclaw chat")}                   ${chalk.dim("# interactive REPL - talk to your agent")}`);
     console.log("");
     console.log(chalk.bold("First-time setup:"));
     console.log(`  ${chalk.cyan("bajaclaw setup --interactive")}    ${chalk.dim("# name your agent, set tone, topics, don'ts")}`);
@@ -368,7 +368,7 @@ async function maybeShowWelcome() {
         return;
     // Only show when stdout is a TTY. npm captures postinstall output,
     // so firing the welcome during `npm install` prints to the void and
-    // then marks done — defeating the point. Non-TTY: silent no-op, don't
+    // then marks done - defeating the point. Non-TTY: silent no-op, don't
     // even mark done, so the next interactive run still gets the welcome.
     if (!process.stdout.isTTY)
         return;
@@ -392,7 +392,7 @@ program.hook("postAction", async () => {
         return;
     await maybeNoticeAtExit();
 });
-// Run the first-run welcome before command dispatch. Non-blocking —
+// Run the first-run welcome before command dispatch. Non-blocking -
 // a failure here should never prevent the user's command from running.
 await maybeShowWelcome().catch(() => { });
 // Await the parse so long-running interactive commands (like `chat`)

package/dist/commands/channel.js

@@ -4,7 +4,7 @@ export async function cmdAdd(profile, kind, token, channelId, userId) {
     const cfg = loadConfig(profile);
     // For telegram: `--channel-id` is the user's numeric Telegram id
     //   (the same thing @userinfobot returns). It lives in the allowlist,
-    //   not channelId — telegram adapters route replies by chat id from
+    //   not channelId - telegram adapters route replies by chat id from
     //   the inbound message, not a pre-set channel.
     // For discord: `--channel-id` is the discord channel id; `--user-id`
     //   is the sender to allow. Without a user id, no allowlist is
@@ -25,7 +25,7 @@ export async function cmdAdd(profile, kind, token, channelId, userId) {
     saveConfig(cfg);
     console.log(chalk.green(`✓ added ${kind} channel to ${profile}`));
     if (kind === "telegram" && entry.allowlist?.length === 0) {
-        console.log(chalk.yellow("  note: no user id provided — allowlist is empty (any user can message)"));
+        console.log(chalk.yellow("  note: no user id provided - allowlist is empty (any user can message)"));
     }
 }
 export async function cmdRemove(profile, kind) {