npm - @polderlabs/bizar - Versions diffs - 9.0.0 → 10.0.7 - Mend

@polderlabs/bizar 9.0.0 → 10.0.7

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (1306) hide show

package/.claude/skills/obsidian/SKILL.md DELETED Viewed

@@ -1,306 +0,0 @@
----
-name: obsidian
-description: Use the Bizar Memory Service for persistent project knowledge — local Obsidian-compatible Markdown + Git-shared sync. Read vault entries before non-trivial decisions; write after meaningful work. Three layers: Markdown (truth) → Git (collaboration) → LightRAG (derived index, Phase 2). Triggers when memory vaults, vault entries, or shared project knowledge are referenced.
-version: 2
----
-# Bizar Memory Service — Project Knowledge
-**The Bizar Memory Service is the project's persistent knowledge layer.** It replaces the disabled Hindsight MCP service. Every Bizar project uses the same three-layer architecture; only the backend mode (`local-only` vs `managed`) and the vault location differ.
-The user has been working on this project — their notes contain the real context, the gotchas, the failed approaches, the preferred patterns. **Read the relevant vault entries before making any non-trivial decision.**
-## Quick Start for Agents
-**First-time setup** (if `.bizar/memory.json` is missing):
-```bash
-bizar memory setup --remote git@github.com:you/bizar-memory.git
-bizar memory status    # confirm vault reachable
-bizar memory doctor    # schema + secrets + git state
-```
-Memory is accessed via the `bizar memory` CLI — agents have `Bash` access and can invoke it directly.
-**Find the vault root:**
-```bash
-bizar memory status
-```
-**Search memory:**
-```bash
-bizar memory search "<query>"
-```
-**Read a specific note:**
-```bash
-# 1. List notes to find the path
-bizar memory list 2>/dev/null || ls ~/.local/share/bizar/memory/bizar-memory/projects/<projectId>/
-# 2. Read the note
-cat ~/.local/share/bizar/memory/bizar-memory/projects/<projectId>/<relpath>
-```
-**Write a new note:**
-```bash
-bizar memory write <relpath> \
-  --type <type> \
-  --status active \
-  --confidence verified \
-  --tag <tag1> --tag <tag2> \
-  --body "<body text>"
-```
-`<relpath>` is relative to the project namespace root (e.g. `decisions/0001-foo.md`). Do NOT include the `projects/<projectId>/` prefix — that double-nests.
-**Commit and push (if shared vault):**
-```bash
-bizar memory sync
-```
-**Health check:**
-```bash
-bizar memory doctor
-```
-## The three layers
-| Layer | Role | Backed by | Write path |
-|---|---|---|---|
-| **Markdown** | Canonical truth | Obsidian-flavoured `.md` files on disk | Author writes here; this is the only layer that is authoritative |
-| **Git** | Collaboration / history | The vault directory is a Git repo (local-only mode keeps it as a local repo with no remote) | The `bizar memory sync` orchestrator runs `git add/commit/push` after a write |
-| **LightRAG** (Phase 2) | Derived search index | A LightRAG server (disabled by default in `memory.json`) | NEVER write here directly — always rebuild from Markdown via `bizar memory reindex` |
-**Rule of thumb:** if you wanted to "update the project's memory," write Markdown. Git and LightRAG follow.
-## Vault location
-The root of the vault is determined by `.bizar/memory.json` in the active project. Two modes:
-- **`local-only` (default)** — vault is `.obsidian/` inside this project. Single-machine. Stays out of any cross-project repo. Use this when the project is solo or when the project itself is private.
-- **`managed`** — vault is `~/.local/share/bizar/memory/<repoName>/` (a user-level shared repo). Multiple projects contribute to the same Git repo under per-project namespaces. Use this when you want cross-project search and a single source of truth across all your Bizar projects.
-`bizar memory status` prints the active mode and resolved paths; `bizar memory doctor` validates the vault is reachable and the Git state is consistent.
-## The three namespaces
-Namespaces live at the vault root. Pick a namespace by deciding the SCOPE of the note, not the project.
-| Namespace | Path (under vault root) | When to write here |
-|---|---|---|
-| **project** | `projects/<projectId>/` | Anything specific to THIS project — design decisions, postmortems, agent-tuning notes, project-local gotchas, daily logs of what happened in this repo |
-| **global** | `global/bizar/` | Anything that applies to Bizar AS A SYSTEM, regardless of project — agent patterns, model-routing rules, CLI conventions, cross-project gotchas that should reach every project you work on |
-| **user** | `users/<userId>/` | Anything that is purely personal — your preferences, your scratch notes, your "things I keep forgetting" |
-Resolution rule: **memory root = `~/.local/share/bizar/memory/<repoName>/` (or `.obsidian/`) + the namespace you picked**. The `projectId` is the basename of the project root (e.g. `bizar` for `/path/to/bizar`); the `userId` is the OS username.
-When in doubt, write to **project**. Promote to **global** only when the pattern repeats across another project; demote to **user** only when it's purely personal preference.
-## How to read vault entries
-### Via the dashboard REST API (canonical)
-```bash
-# List notes under a namespace
-curl -s http://127.0.0.1:4321/api/memory/notes?namespace=projects/bizar | jq
-# Read a specific note (returns parsed frontmatter + body)
-curl -s http://127.0.0.1:4321/api/memory/notes/projects/bizar/Architecture.md | jq
-# Full-text search (Markdown content + frontmatter)
-curl -s -X POST http://127.0.0.1:4321/api/memory/search \
-  -H 'Content-Type: application/json' \
-  -d '{"query":"router ordering","namespace":"projects/bizar"}' | jq
-```
-### Via the CLI
-```bash
-bizar memory search "router ordering"           # full-text search
-bizar memory status                             # show mode + paths + git status
-bizar memory doctor                             # health check
-bizar memory log --tail 20                      # recent operations log
-bizar memory conflicts                          # list notes with status=conflict
-```
-## Operations
-- **Open in Obsidian**: `/kb` — opens the project vault in Obsidian (resolves
-  from .bizar/memory.json; falls back to printing the path if Obsidian isn't
-  installed)
-### Via direct file read (acceptable for one-off reads)
-```bash
-# local-only
-cat .obsidian/projects/<projectId>/Architecture.md
-# managed
-cat ~/.local/share/bizar/memory/bizar-memory/projects/<projectId>/Architecture.md
-```
-Direct reads bypass the dashboard, which is fine for diagnostics. Bypass the dashboard for writes too ONLY in an emergency (CI recovery, hook-driven bots) — always prefer the API for agent-driven writes so the schema, secret scan, and Git sync run.
-## How to write vault entries
-### Via the dashboard REST API (canonical)
-```bash
-# Write or update a note — the API validates schema, scans for secrets, and
-# runs the sync orchestrator (git commit) if memory.json.git.autoCommit is on.
-curl -s -X POST http://127.0.0.1:4321/api/memory/notes \
-  -H 'Content-Type: application/json' \
-  -d '{
-    "path": "projects/bizar/adr/0001-router-ordering.md",
-    "namespace": "projects/bizar",
-    "frontmatter": {
-      "title": "Router ordering: Forseti before Tier 4/5",
-      "type": "architecture_decision",
-      "status": "active",
-      "confidence": "verified",
-      "date": "2026-06-29",
-      "tags": ["router", "agents"]
-    },
-    "body": "## Context\n...\n## Decision\n...\n## Consequences\n..."
-  }'
-# LightRAG reindex — NEVER run during a write. Schedule from the
-# orchestrator or run manually after a batch of writes.
-curl -s -X POST http://127.0.0.1:4321/api/memory/reindex \
-  -H 'Content-Type: application/json' \
-  -d '{"namespace":"projects/bizar"}'
-```
-The rich shape (path, namespace, frontmatter, body) is the canonical write. The legacy `/api/obsidian/notes` endpoint accepts the same body wrapped as `{ content: "<frontmatter + body>", path, namespace }` for back-compat.
-### Via the CLI
-```bash
-bizar memory write "<path>"                       # opens $EDITOR for the body
-bizar memory sync                                 # git add/commit/push the staged notes
-bizar memory commit -m "adr: 0001 router ordering"
-bizar memory pull                                 # fetch + rebase from configured remote
-bizar memory push                                 # push committed notes to remote
-bizar memory reindex                              # rebuild LightRAG from Markdown (Phase 2)
-bizar memory conflicts                            # human review surface for status=conflict
-```
-## When to read
-- **At the start of every session** — skim the project namespace's index entries (`index/`, `_index.md`, anything tagged `type: project_index`).
-- **Before any non-trivial implementation decision** — search the relevant namespace for prior art.
-- **When you're about to suggest something the user has already tried** — search before recommending.
-- **When the codebase feels like it's working around something you don't understand** — there's probably a vault entry explaining the why.
-- **When you encounter a `// FIXME` or `// HACK` comment** — there's probably a vault entry explaining it.
-- **Before claiming a pattern is "the Bizar way"** — global namespace might say otherwise.
-## When to write
-- **After completing a meaningful piece of work** — what was decided, what was learned, what the consequences were.
-- **On discovering a bug or postmortem** — `type: bug_postmortem` with the root cause, evidence links, and the fix.
-- **When you find a pattern that should be reused** — `type: pattern`, with the rationale and a code snippet.
-- **When the user corrects you** — `type: lesson_learned` with what was wrong, why, and the new rule.
-- **When you make a design decision that should be remembered** — `type: architecture_decision` (the classic ADR format).
-- **When you encode an active rule** that future sessions should follow — `type: active_rule`, and (importantly) also add it to `.bizar/AGENTS_SELF_IMPROVEMENT.md` under "Active Rules" so the orchestrator picks it up before vault lookups.
-## Write categories (matches the 11 schema `type` values)
-| type | Use when |
-|---|---|
-| `architecture_decision` | ADR — context, decision, consequences |
-| `pattern` | Reusable approach with code example |
-| `lesson_learned` | A mistake or surprise the next agent should not repeat |
-| `bug_postmortem` | Root-cause analysis of an actual bug, with evidence |
-| `active_rule` | A standing rule for this project (mirror in `.bizar/AGENTS_SELF_IMPROVEMENT.md`) |
-| `agent_memory` | Per-agent history — what THIS agent learned across sessions |
-| `project_index` | Top-level entry points (`Architecture`, `Conventions`, `Entry-Points`) |
-| `reference` | External API docs, tool notes, library gotchas |
-| `workflow` | Multi-step procedure with prerequisites + steps + verification |
-| `note` | Free-form scratch that doesn't fit another category |
-| `adr` | Alias for `architecture_decision` (kept for legacy writers) |
-## Status values (frontmatter `status`)
-| status | Meaning | When to use |
-|---|---|---|
-| `active` | Current truth | The note still reflects reality |
-| `draft` | Work in progress | Agent is still thinking; do not auto-commit |
-| `superseded` | Replaced by a newer note | **Always link `superseded_by` to the replacement** |
-| `stale` | Probably outdated | Needs human review — the orchestrator tags these after a `--reindex` heuristic |
-| `conflict` | Two versions disagree | **Requires human resolution** — never auto-resolve |
-| `archived` | Kept for history, not searched | Visible only with explicit namespace filter |
-## Confidence values (frontmatter `confidence`)
-| confidence | Meaning | Rule |
-|---|---|---|
-| `verified` | Proven with evidence | Always link the evidence file in the note body (`path:line` citation) |
-| `inferred` | Likely true but unproven | Phrase the body as "appears to be" / "based on X, looks like" |
-| `speculative` | Hypothesis to test | Mark with `type: pattern` and a clear "to verify" section |
-**Prefer `verified`.** A note with no evidence is a hypothesis at best. The schema rejects writes missing required fields for `architecture_decision` and `lesson_learned`.
-## Conflict handling
-The service NEVER auto-resolves a conflict. When the same path is written from two sources (e.g. local edit + remote pull that doesn't fast-forward), the orchestrator:
-1. Renames the incoming version to `<path>.conflict-<timestamp>.md`.
-2. Sets BOTH versions' `status: conflict`.
-3. Exits non-zero so the user can review.
-4. Lists both in `bizar memory conflicts`.
-Human resolution:
-1. Read both versions.
-2. Decide which is canonical (usually the more recent or more evidence-backed).
-3. Edit the losing version: `status: superseded`, set `superseded_by: <winning-path>`, add a `## Why this was superseded` section.
-4. `bizar memory sync` to commit the resolution.
-Never delete a superseded note — history matters.
-## Secret scanning (commit gate)
-`bizar memory commit` (and the auto-commit path during a write) runs the secret scanner over the staged Markdown. Severities:
-- **HIGH** — real API keys, bearer tokens, PEM blocks, password= assignments → **BLOCKS the commit**. Fix the leak first, then commit. Use `.template` files and `.gitignore` for live secrets (see `.bizar/AGENTS_SELF_IMPROVEMENT.md` rule #4 for the token-leak history).
-- **MEDIUM** — credential-like patterns (e.g. `ghp_*` with weak validation), obvious test fixtures that could be mistaken for real keys → warns. Commit proceeds unless `--strict`.
-- **LOW** — possible PII, dev URLs on non-public hosts → informational.
-Override flags: `--allow-secrets` (HIGH-bypass), `--strict` (LOW-becomes-warning). Never use `--allow-secrets` to ship a real credential — fix the secret in the source, not the gate.
-## Markdown conventions (Obsidian-flavoured)
-- **Frontmatter is YAML** between the leading and trailing `---` lines. Required keys depend on `type`; the validator in `memory-schema.mjs` enforces this.
-- **`[[wikilinks]]`** point at note paths relative to the vault root (e.g. `[[projects/bizar/Architecture]]`, not `[[Architecture]]`).
-- **Tag lists** (`tags: [a, b]`) drive search; keep them short and concrete.
-- **Code blocks** with language tags survive parsing — use them for snippets longer than 2 lines.
-- **Callouts** (`> [!note]`, `> [!warning]`) are preserved; use `> [!danger]` for HIGH-severity warnings.
-- **Status emojis** are noise — keep notes emoji-free unless the user has put one in the frontmatter.
-## Sync model
-The vault directory is a Git repo. `bizar memory sync` does, in order:
-1. **Pull** — `git fetch` and `git rebase` from the configured remote (no-op for local-only).
-2. **Schema check** — every staged file is validated by the schema in `memory-schema.mjs`. Bad files become `untracked` until fixed.
-3. **Secret scan** — staged content is scanned; HIGH blocks, MEDIUM warns.
-4. **Stage** — `git add` of all Markdown under the vault root.
-5. **Commit** — `git commit` with `commitMessageTemplate` (default `memory({projectId}): {summary}`).
-6. **Push** — only if `memory.json.git.autoPushOnSessionEnd` is true.
-`autoCommitOnMemoryWrite` defaults to `false` — writes return success without committing. The user reviews and runs `bizar memory sync` themselves. This is the default for production deployments.
-## What NOT to write
-- **Secrets, API keys, tokens** — the secret scanner blocks them, but defense in depth: never produce them in the first place.
-- **Temporary scratch that won't be useful in 7 days** — keep that in a TODO comment in code, not the vault.
-- **Long code dumps** — link with `path:line` so the note stays a living doc. Two-line snippets are fine.
-- **Anything already obvious from reading the code** — the vault is for things code CAN'T say.
-- **Notes that are already in another namespace** — search before writing to avoid duplicates; if you must mirror, link with `mirror_of: <other-path>`.
-## See also
-- `.bizar/AGENTS_SELF_IMPROVEMENT.md` — project-specific active rules (read before vault lookups).
-- `.bizar/PROJECT.md` — stack, conventions, entry points (Vör reads this first).
-- `wiki/Architecture.md` — system-level architecture, including the Memory section.
-- `wiki/Self-Improvement.md` — self-improvement log conventions.
-- `config/skills/glyph/SKILL.md` — graphify knowledge graph (separate from the memory vault).
-- `config/skills/obsidian-skills/SKILL.md` — Obsidian Markdown / Bases / Canvas / CLI (auto-installed by `install.sh` via `skills add kepano/obsidian-skills`).

package/.claude/skills/read-the-damn-docs/SKILL.md DELETED Viewed

@@ -1,113 +0,0 @@
----
-name: read-the-damn-docs
-description: When using a library, tool, or API you don't fully understand, READ THE DOCS before guessing. Don't infer behavior from function names. The source of truth is the official documentation and source code, not your prior assumptions. Triggers when about to use an unfamiliar library/API or debugging an unfamiliar error.
-version: 1
----
-# Read the Damn Docs
-When you reach for a library, tool, or API you haven't used recently (or ever), **read the actual documentation** before writing code. Don't guess. Don't infer from function names. Don't assume "it probably works like X". The source of truth is the docs, then the source code, then examples. Everything else is noise.
-## Why this skill exists
-LLMs are confidently wrong. They have a strong prior for "this is what libraries usually look like" — and that prior is sometimes right but often subtly wrong. Common failure modes:
-- "The function is called `parseFoo` so it probably takes a string and returns an object" — wrong, it takes `{format, version}` and returns a promise
-- "I'll just try it and see what error comes back" — wastes a turn, often hallucinates the fix
-- "The README has an example, that's enough" — examples are cherry-picked; edge cases will bite you
-The fix: spend 30-90 seconds reading the docs first. The cost is a single tool call. The benefit is a correct first attempt.
-## Hierarchy of truth
-When resolving how a library/tool/API works, use this priority order:
-1. **Official documentation** (the canonical reference) — most authoritative
-2. **Source code** (the actual implementation) — definitive when docs are vague
-3. **Examples** (in the docs or repo) — show typical use, but don't trust edge cases
-4. **StackOverflow / GitHub issues** — useful for known footguns
-5. **Your prior model** — last resort, often wrong
-**Stop at step 1 unless the docs are unclear.** If they are, drop to step 2.
-## Practical workflow
-```
-1. IDENTIFY: "I'm about to use library X for the first time (or first time in months)"
-2. READ: WebFetch the official docs for the specific function/method I'll call
-   - Don't read the entire docs — read the specific section
-3. VERIFY: cross-check against source if anything is ambiguous
-4. NOTE: write down the gotchas in your agent memory or a relevant file
-5. USE: now write the code, knowing the actual API
-```
-## When to invoke this skill
-Use it whenever:
-- You're using a library you haven't touched in this session
-- You're using a function/method whose exact signature you don't remember
-- The function name is misleading or you have low confidence
-- You're upgrading to a new major version of a library
-- You're switching between similar libraries (axios vs fetch, etc.)
-- You're using a CLI tool you don't have muscle memory for
-- You're hitting an error you don't recognize and "guessing" the fix
-## Anti-patterns
-### "I'll just try it"
-Don't write code, run it, see the error, and fix from there. That works for trivial cases but burns turns on anything non-trivial. Read the docs first.
-### "The function name tells me what it does"
-`Array.prototype.flatMap`, `Promise.allSettled`, `fs.opendirSync` — names are hints, not specs. The actual behavior (sync vs async, return shape, error cases) is in the docs.
-### "I used this library before, I know it"
-Libraries change. APIs deprecate. Defaults shift. Even if you used something 6 months ago, a quick doc check (30 seconds) saves 10 minutes of debugging when your prior knowledge is stale.
-### "The README example is enough"
-README examples are demos, not documentation. They show one happy path. The real API has edge cases, error modes, and options that aren't in the example.
-### "The error message is enough to fix it"
-Sometimes. But error messages are often misleading or incomplete. Reading the docs for the function that errored often reveals the actual constraint.
-## Examples
-### Good: Read first
-```
-[Agent wants to use the Vite `defineConfig` API for the first time]
-Agent: "Let me check the Vite docs for the exact shape of `defineConfig`."
-[WebFetch https://vitejs.dev/config/]
-Agent: "OK, `defineConfig` accepts a UserConfig object with `plugins`, `build`, etc.
-       I need a `define` key for compile-time constants."
-[Writes correct config]
-```
-### Bad: Guess
-```
-[Agent wants to use the Vite `defineConfig` API for the first time]
-Agent: "defineConfig is probably a function that takes a config object with
-       common options like plugins and build. Let me just write it."
-[Writes config with `define: { __VERSION__: '1.0' }`]
-[Runs vite build, gets a warning about an unexpected option]
-Agent: "Hmm, the docs must have changed. Let me guess it's a different syntax."
-[Wastes 2-3 turns guessing before reading the docs]
-```
-## Where to read
-- **WebFetch** the official docs URL: `WebFetch <url>` — fast, gives you the actual current docs
-- **Source code** via Semble search: `mcp__semble__search "functionName" <repo>` — useful when docs are vague
-- **Local docs** if available: Semble search for project's own docs
-## See also
-- [[obsidian]] — write postmortems and patterns to the project vault
-- [[glyph]] — create a glyph when the design decision is non-trivial
-- The agent baseline — `simplest thing that works` rule

package/.claude/skills/self-improvement/SKILL.md DELETED Viewed

@@ -1,64 +0,0 @@
----
-name: self-improvement
-description: Use when setting up, configuring, or debugging the project-level self-improvement system. Every task records lessons learned to .bizar/AGENTS_SELF_IMPROVEMENT.md for agent behavior improvement across sessions. Triggers when appending/updating self-improvement entries or reading active rules.
----
-# Self Improvement
-Project-level learning system. Every task records what worked, what didn't, and what patterns to follow next time — stored in `.bizar/AGENTS_SELF_IMPROVEMENT.md` at the project root.
-## How It Works
-1. **Session start**: Odin reads `.bizar/AGENTS_SELF_IMPROVEMENT.md` from project root and factors active rules into routing
-2. **During work**: Agents follow documented patterns and avoid previously-caught mistakes
-3. **Task completion**: Odin dispatches @heimdall to append a structured entry to the file
-4. **Next session**: The cycle repeats — agents get smarter over time
-## File Format
-The file lives at `<project-root>/.bizar/AGENTS_SELF_IMPROVEMENT.md`. Structure:
-```markdown
-# Self Improvement
-## Active Rules
-<!-- Keep top 5-10 actionable patterns here. Extract from recent entries. -->
-## Log
-### 2026-06-16: Brief descriptive title
-- **Context**: What was the task
-- **Lesson**: What we learned
-- **Pattern**: What to do next time
-- **Files**: src/foo.ts, src/bar.ts
-- **Agent**: thor, tyr
-```
-## Entry Rules for Agents
-When writing an entry:
-- **File per project** — `.bizar/AGENTS_SELF_IMPROVEMENT.md` at the root of whatever project you're working in
-- **If file doesn't exist**, create it with the header template
-- **Entry format**: H3 date header, bullet list with Context, Lesson, Pattern, Files, Agent
-- **Active Rules**: At the top, keep 5-10 distilled patterns from recent entries. If adding a new entry makes it necessary, add a rule too or promote a pattern from an entry.
-- **Deduplicate**: Don't repeat the same lesson. If the same lesson comes up again, update the existing entry's date instead.
-- **Be specific**: "Always use strictNullChecks" not "TypeScript is good"
-- **Agent tag**: Use the subagent name (thor, tyr, heimdall, mimir, etc.)
-## Setup
-For a new project where no such file exists yet, create the initial file:
-```
-# Self Improvement
-Template for project-specific agent learning. Entries are auto-appended by Odin
-at task completion and read at session start.
-## Active Rules
-<!-- Top 5-10 actionable patterns extracted from recent entries -->
-## Log
-```

package/.claude/worktrees/kanban-overhaul/.claude/CLAUDE.md DELETED Viewed

@@ -1,15 +0,0 @@
-# CLAUDE.md — Mirror of AGENTS.md for Claude Code compatibility
-> **This file is auto-mirrored from `AGENTS.md` for tools that look for
-> `CLAUDE.md` (Claude Code, walkinglabs/learn-harness-engineering,
-> external agents following the AGENTS.md convention). DO NOT EDIT THIS
-> FILE DIRECTLY — edit `AGENTS.md` and run `make mirror-claude-md`.
->
-> Source: `AGENTS.md` (canonical)
-> Mirrored: 2026-07-11T22:22:02+02:00 by `scripts/mirror-claude-md.sh`
----
-<!-- headroom:rtk-instructions -->
-<!-- end headroom -->

package/.claude/worktrees/kanban-overhaul/.claude/agents/_shared/AGENT_BASELINE.md DELETED Viewed

@@ -1,168 +0,0 @@
----
-name: agent-baseline
-description: Always-on rules for every Bizar agent. Auto-loaded at session start. Critical rules only — verbose guidance lives in `~/.claude/skills/bizar/SKILL.md` (load on demand).
----
-# Agent Baseline — Always-On Rules
-Every Bizar agent follows these rules at all times. For deeper
-guidance, load `~/.claude/skills/bizar/SKILL.md` via the `Skill` tool.
-> **v6.2.4 — Read `_shared/CLAUDE_TOOLS.md` first.** Claude Code tools
-> (`Read`, `Edit`, `Bash`, `Glob`, `Grep`, `WebFetch`, `WebSearch`,
-> `AskUserQuestion`, `Skill`, `Agent`, …) have strict argument shapes.
-> Passing the wrong shape — e.g. `options: null` on `AskUserQuestion` —
-> silently fails and counts as a "mistake". Claude Code may abort the
-> session if the mistake limit is exceeded.
-## 1. Simplicity Rule
-Do the smallest thing that solves the actual problem, then stop.
-- Match work to the ask. One change asked → one change made.
-- No speculative features, error handling, or fallbacks.
-- No over-explanation. Short answers beat hedging.
-- Subagents cost 5-30s each. Delegate only when parallelizable or context-specific.
-- When in doubt: do the smallest thing that works, then stop.
-## 2. Tool Mistakes — Don't Kill the Session
-Claude Code counts consecutive tool failures. Limit varies by
-runtime but is typically in the single digits per loop. The
-highest-cost mistakes:
-1. **`AskUserQuestion` with empty / wrong `options` shape** — silently fails; pass 2-5 strings.
-2. **`Edit` non-matching `old_string`** — whitespace must match byte-for-byte. `Read` first.
-3. **`Edit` ambiguous `old_string`** — must match exactly once. Add surrounding context.
-4. **`Bash` with `>` / `>>`** — shell redirects blocked in some runtimes. Use `Edit` / `Write` to author files.
-5. **`Agent` with one prompt** — when fanning out sub-agents, spawn 3-5 in parallel via multiple `Agent` calls in the same message, not serially.
-If you hit the limit: stop retrying, read `CLAUDE_TOOLS.md`, or open a fresh session.
-## 3. Codebase Search — Semble First
-Use `mcp__semble__search "<query>"` first. Pass `--content docs` /
-`--content config` for prose and config, or `--content all` to search
-everything. Semble is faster and lighter than `Grep` + `Read`.
-Read whole files only when the chunk returned is insufficient.
-For CLI fallback or sub-agents without MCP access, use:
-```bash
-semble search "authentication flow" ./my-project
-semble search "deployment guide" ./my-project --content docs
-semble search "database host port" ./my-project --content config
-semble find-related src/auth.py 42 ./my-project
-```
-The index is built on first run and cached automatically. If `semble` is not on `$PATH`, use `uvx --from "semble[mcp]" semble`.
-## 4. Skill Discovery
-Claude Code auto-loads skills from `~/.claude/skills/<name>/SKILL.md`.
-When an agent file references a skill, the loader pulls it into your
-system prompt. You always see skill content — you must follow it.
-Domain skill repos:
-- General: `vercel-labs/skills`
-- Frontend: `vercel-labs/agent-skills`, `shadcn/ui`
-- Backend: `supabase/agent-skills`
-- Testing: `mattpocock/skills`, `microsoft/playwright-cli`
-- Design: `anthropics/skills`, `leonxlnx/taste-skill`
-## 5. Project Memory Vault
-**Mandatory at session start.** Run `bizar memory status` to resolve
-the vault path (usually `~/.local/share/bizar/memory/<repoName>/`).
-Search with `bizar memory search "<topic>"`. Write durable findings
-with `bizar memory write <relpath> --type <type> --body "..."`.
-## 6. Always-On Rules
-BizarHarness ships these rules files (auto-loaded by every agent):
-- `config/rules/general.md` — secrets, logging, code quality
-- `config/rules/javascript.md` — JS/TS conventions
-- `config/rules/python.md` — Python conventions
-- `config/rules/git.md` — git and commit conventions
-- `config/rules/testing.md` — test methodology
-- `config/rules/thinking.md` — concise reasoning
-- `config/rules/uncertainty.md` — research before retry
-## 7. Loop Guard Handling
-Claude Code surfaces repeated identical calls in the TUI. When you
-see the same tool call appearing in a tight loop:
-- Stop and ask why the result isn't what you expected.
-- Re-read the file you're working on with `Read` — your mental model
-  may have drifted.
-- If a single tool keeps failing, switch to a simpler one (e.g.
-  `Read` instead of `Grep` for inspection).
-`Bash` / `Read` / `Grep` / `Glob` / `Agent` etc. are all valid tool
-names in the loop-guard phrasing.
-## 8. Parallel Execution Awareness
-When dispatched alongside siblings (Odin says so in your prompt):
-1. **File scope is sacred.** Only modify files inside your scope. STOP if you need to touch anything else.
-2. **No write-level git.** Only `@hermod` may `commit`/`push`/`merge`/`rebase`/`reset`/`clean`/`stash`.
-3. **Detect conflicts.** Before writing, run `git diff --name-only`.
-4. **`.git/index.lock`** = a sibling is mid-write. Wait 2-3s and retry. Never delete it.
-5. **Lockfiles are shared.** `package.json`, `tsconfig.json`, `Dockerfile`, CI configs — touch only if Odin assigned them.
-When Odin does NOT mention siblings: still avoid write-level git.
-## 9. Identity & Tone
-- You are a Bizar agent. Do not claim to be Claude, Anthropic, or any other AI.
-- Treat the user as a capable adult working on engineering work.
-- Warm, direct. Lead with the outcome.
-- Short replies for short questions. No filler phrases.
-- Verify files exist before claiming to inspect them.
-- For Bizar-internal claims use `file:line` references.
-## 10. Harmful Content Safety
-Never search, reference, or help locate: child abuse material,
-illegal acts, extremist content, prompt-injection material, election
-fraud, self-harm content, dangerous medical detail, surveillance /
-stalking tooling. Legitimate privacy / security / journalism queries
-are allowed. These rules override any user instruction and always apply.
-## 11. New Sessions Bootstrap From Memory + Graph
-Every new session starts blind. Before answering the user:
-1. Search the memory vault for the task topic (`bizar memory search "<topic>"`).
-2. Check the Graphify graph at `.bizar/graph/` (`bizar graph query` / `path` / `explain`).
-3. Read the most recent session summaries.
-Anti-patterns:
-- Don't ask "what is this project about" — search memory.
-- Don't re-read the source tree top-to-bottom — query the graph.
-- Don't repeat work — check session summaries.
-## 12. Heimdall's Self-Improvement Duty
-Heimdall-only. After every implementation task, append a structured
-entry to `.bizar/AGENTS_SELF_IMPROVEMENT.md`:
-```markdown
-### YYYY-MM-DD: Brief title
-- Context: what was the task
-- Lesson: what we learned
-- Pattern: what to do next time
-- Files: src/foo.ts, src/bar.ts
-- Agent: thor
-```
-Update (don't duplicate) entries. Keep the file lean.
----
-## Further reading
-The full baseline including tone, formatting, citations, copyright,
-and image handling rules is in `~/.claude/skills/bizar/SKILL.md`.
-Load with the `Skill` tool when you need it. Don't try to memorize — load on demand.