Skill Claude CodeCodex
Rules for writing, editing, and retiring skill and workflow files for LLM agents. Use when creating a SKILL.md or workflow document, reviewing one, adding a rule after an incident, or deciding whether a skill should exist at all.
Skill Claude CodeCodex
Entry point for any DeepReason problem. Routes work to exactly one subskill at a time and enforces the scope contract. Use when asked to manage, diagnose, or fix anything in DeepReason.
Skill Claude CodeCodex
Question discipline for working DeepReason - route every question to the cheapest authority (record, then framework, then operator), translate the operator's shorthand into typed obligations, and frame uncertainty as falsifiable forks. Load whenever an operator message is ambiguous or short, whenever any workflow…
Skill Claude CodeCodex
Audit dimension - find broken code by running every committed verification instrument and tabling deltas against the recorded baselines. Invoked by dr-audit-orchestrator when broken.md is missing.
Skill Claude CodeCodex
Audit dimension - find code that is no longer used, via a reference census over every top-level symbol, one package per invocation. Invoked by dr-audit-orchestrator when dead.md is missing or has pending package rows.
Skill Claude CodeCodex
Audit dimension - find where documentation deviates from code, using the executable doc checks plus a staleness and unchecked-claims census. Invoked by dr-audit-orchestrator when docs-drift.md is missing.
Skill Claude CodeCodex
Audit dimension - verify every ledgered operator design law is actually enforced somewhere in code or gates, so stated goals and shipped behavior cannot silently diverge. Invoked by dr-audit-orchestrator when goal-trace.md is missing.
Skill Claude CodeCodex
Router for the code-audit family. Runs five read-only audit dimensions (broken, dead, docs-drift, spec-drift, goal-trace) over the repo and produces AUDITREPORT.md plus ready-to-send fix prompts. Use when the operator asks for an audit, a sweep, or "what is broken / unused / out of date". Findings only — fixes happen…
Skill Claude CodeCodex
Audit dimension - find where the code deviates from the spec series and where shipped surface is spec-silent, by a two-direction term census. Invoked by dr-audit-orchestrator when spec-drift.md is missing.
Skill Claude CodeCodex
Record an operator's suggested change verbatim and split it into numbered requirements (REQUEST.md). First phase of every change tranche; no interpretation, no code reading.
Skill Claude CodeCodex
Entry point for implementing an operator-suggested change to DeepReason. Routes through capture, spec, plan, stepwise execution, validation, and delivery — one phase at a time, with a verbatim request ledger nothing may drift from. Use when the operator says "add", "change", "make it", "I want", or suggests any…
Skill Claude CodeCodex
Close a validated change tranche — final commit and push, requirement-by-requirement reconciliation against the operator's verbatim words, and the delivery report (DELIVERY.md). Use only after VALIDATION.md says PASS.
Skill Claude CodeCodex
Locate the cause of a DeepReason defect from the typed record, not from code reading. Produces DIAGNOSIS.md naming one primary cause with evidence pointers. Use only after GOAL.md exists.
Skill Claude CodeCodex
The driving manual for DeepReason - how to run the harness properly (session preflight, the public CLI lifecycle, live-run ladders) and where to look before modifying anything or when diagnosing a problem. An index over the owning authorities (CLAUDE.md, docs/map, the workflow skills), not a replacement for them. Load…
Skill Claude CodeCodex
Execute exactly one unchecked step from CHECKLIST.md, prove its done-criterion, record the output, and stop. The only skill in the change workflow allowed to modify the tree. Invoke repeatedly, once per step.
Skill Claude CodeCodex
Communication discipline for EVERY message the operator will read - intermediary status reports, phase-boundary reports, STOP questions, failure notices, and the final message alike. Answer the operator's actual worry first, gloss every technical term conservatively in plain language as you go, and close every FINAL…
Skill Claude CodeCodex
Apply an approved FIX.md to DeepReason with regression tests and the full gate. The only skill allowed to modify production code. Use after FIX.md passes its approval gate.
Skill Claude CodeCodex
Convert an approved SPEC.md into an ordered, checkable step list (CHECKLIST.md) where each step has one done-criterion. Use after SPEC.md, and again (partial re-plan) after a validation failure.
Skill Claude CodeCodex
Design the smallest correct DeepReason fix for a reproduced cause, as FIX.md. No code changes yet. Use after REPRO.md confirms the diagnosis.
Skill Claude CodeCodex
Demonstrate a diagnosed DeepReason cause with the smallest offline artifact — a failing test or a minimal script against the committed record. Use after DIAGNOSIS.md, before any fix.
Skill Claude CodeCodex
Turn a vague DeepReason problem statement into one bounded, falsifiable tranche goal (GOAL.md). Use at the start of every tranche, before any diagnosis or code reading.
Skill Claude CodeCodex
Translate a captured request into a concrete, bounded change specification (SPEC.md) with per-requirement acceptance checks and recorded assumptions. Use after REQUEST.md exists or gains amendments.
Skill Claude CodeCodex
Prove the completed change against every acceptance check in SPEC.md and the full DeepReason gate, producing VALIDATION.md. Use when every CHECKLIST.md step is checked. Validates only; never patches.
Skill Claude CodeCodex
Prove a DeepReason fix against its GOAL.md success criterion, optionally with one guarded live run, and record the honest outcome in VERIFY.md. Use after dr-implement-fix.