Instructions file
Instructions for denn-gubsky/loomcycle, covering claude.md — loomcycle project guide, project context, development workflow, karpathy guidelines (coding discipline) and security rules — non-negotiable.
Agent
Capability bench agent for low-tier (haiku-class) candidate models. Tier slot is the cheapest production workload — schema-correct output, single-shot tool calls, short multi-turn loops. NOT for content quality; that's middle tier.
Agent
Capability bench agent for middle-tier (sonnet-class) candidate models. Tier slot is production content quality — CV rewrites, QA answers, profile enrichment — plus full MCP read/write cycles and self-correction at scale.
Skill Claude CodeCodex
Create and curate graph edges between chunks (and across documents) in a chunked-graph Document.
Skill Claude CodeCodex
Import a Markdown file into a chunked-graph Document — deterministic round-trip vs semantic chunking.
Skill Claude CodeCodex
Safely move, reparent, reorder, promote, or delete chunks in a chunked-graph Document.
Skill Claude CodeCodex
Split arbitrary prose or a Markdown file into a sensible chunk hierarchy in a chunked-graph Document.
Agent
Reviews a slice of a code repository for bugs, security, and quality with confidence-based filtering (reports only high-confidence issues). Read-only.
Skill Claude CodeCodex
How to review a slice of a code repository with high precision — confidence-scored findings, project-guideline compliance, and Karpathy behavioral guidelines (simplicity, surgical scope, surfaced assumptions). Use when reviewing or auditing source files.
Agent
AgentDef runtime smoke test agent. Walks through create/get/list/fork/promote/retire in one run.
Agent
Drains the findings channel and summarises what it sees. v0.8.4 Channel-tool runtime smoke-test agent.
Agent
Publishes structured findings to the findings channel. v0.8.4 Channel-tool runtime smoke-test agent.
Agent
Exercises Context.help index and detail ops in one run.
Agent
Exercises Context tool's four PR-1 ops in one run.
Agent
Submits and queries Evaluations against a previously-run worker's runid.
Agent
Trivial deterministic-output agent. Its runid becomes the eval target.
Agent
You are a no-op memory keeper. Always finish immediately.
Agent
Search memory with dedup and finish.
Agent
Two-run Memory smoke test agent. Writes on the first call, reads on the second.
Agent
You are a no-op agent. Always finish immediately.
Agent
You are a no-op soak agent. Always finish immediately.
Agent
Channel-tool runtime test agent. Publishes to findings with optional deliverat, then subscribes.
Agent
One-turn greeter for the v0.8.2 usertier runtime-fallback test. No tools.
Agent
You are a no-op webhook responder. Always finish immediately.