betmoar

Use when you need to read a large body of code or text fast and cheaply — whole subsystems, sprawling logs, many files — by fanning the work out across parallel glm-code-crawler shards via cc-proxy and merging the digests. You orchestrate the shards; the agent reads one shard each. Requires cc-proxy running. Give it the paths/globs and the crawl question.

Use when a spec or plan has just been written or when the user asks for a second-eyes review of one — convenes a panel of N parallel GLM reviewers, each through a distinct lens, then synthesizes their findings into one scored report. Triggered by the cc-agents PostToolUse hook after a *-design.md (under specs/) or *-plan.md write, or invoked directly. Requires cc-proxy running.

One-time setup for the cc-status plugin, run when the user explicitly invokes /cc-status:setup. Wires ~/.claude/settings.json so the cc-status composer becomes the single statusLine, then seeds ~/.claude/cc-status/segments enabling every discovered plugin segment. Use when the user wants to adopt or enable the cc-status statusline.

How a Claude Code plugin ships a statusline segment that cc-status can compose. Use when developing a plugin that should appear on the cc-status bar — covers the .claude-plugin/statusline.json manifest, the renderer contract (stdin session JSON in, one segment out), ordering, and graceful-empty behavior.

Operate a cc-repete autonomous loop well — write a verifiable mission goal, set up the frozen constitution vs. the evolving brief, author lesson cards, read checkpoints, and tune budgets. Use this whenever the user is starting, configuring, resuming, or debugging a repete run (the /repete, /repete-continue, /repete-status, /repete-cancel commands), mentions a "repete loop", an autonomous/self-evolving loop, a ".repete/" directory, a mission goal, loop checkpoints, context-rot pauses, or a lesson library — even if they don't name the plugin explicitly. The single biggest failure mode is a vague mission goal producing a runaway loop, so reach for this skill before the first iteration runs.

Decide whether and how to run an agent in an autonomous loop — repete (single-session re-inject) vs. ralph (fresh process per iteration) vs. a one-shot, plus how to fight context rot with memory layering. Use this whenever the user is weighing "should I loop this", asks about autonomous/agentic loops, iteration harnesses, context rot, a long unattended run, a ralph loop, set-and-forget agents, or how to keep a long agent run from degrading — even if they don't name a specific tool. Reach for this before architecting a loop, so the choice is grounded in how context rot actually behaves rather than vibes.

One-time setup for the cc-proxy plugin. Writes API keys (GLM_API_KEY, optionally OPENROUTER_API_KEY) to ~/.env, and configures ANTHROPIC_BASE_URL, PROXY_PATH, and the glm-5.2[1m] custom model option in ~/.claude/settings.json so the SessionStart hook can auto-start the proxy and /model can route to GLM. Invoke via /cc-proxy:setup.

Use when the user asks to run as chief operator, orchestrate this, operator mode, or set up an engagement — routing to the operator charter and its start command.

Use at the start of any non-trivial coding, design, refactor, or build task — especially when the request is ambiguous, long-horizon, touches an unfamiliar codebase or domain, or when a previous attempt came back wrong. Do NOT use for small fully-specified edits, single-file mechanical changes, or when the user says "just do it". Use also when a blindspot pass, brainstorm, interview, reference port, implementation plan, implementation notes, pitch doc, or pre-merge quiz is needed.

Keep an ordinary (non-looped) Claude Code session resumable across a context reset — write and refresh the .reload/session.md digest, understand the snapshot -> arm -> rehydrate cycle, and the /clear, /compact, auto-compaction paths. Use whenever a session is getting long, before a /clear or /compact, when context-rot shows (the agent forgetting earlier decisions), or when the user mentions cc-reload, session reload/rehydrate, a ".reload/" directory, a session digest, or losing context on compaction — even if they don't name the plugin. NOT for autonomous loops: if a cc-repete loop is active, that tool owns continuity and cc-reload stands down.

Conventions for reading, writing, validating, and maintaining Open Knowledge Format (OKF) knowledge bundles — directories of markdown concept files with YAML frontmatter, a generated index.md, an append-only log.md, [[wiki-style]] cross-links, and source citations. This skill should be used whenever the user reads, writes, edits, reorganizes, or validates an OKF bundle; encounters an index.md/log.md pair, a concepts/ directory, or .okf markers; or mentions OKF, knowledge bundles, concept notes, or conformance. Apply it even when the user does not say "OKF" explicitly but is clearly working inside a bundle whose project is OKF-activated.