Evidence-driven engineering skills for agents. Smaller changes, stronger proof.
Skill Claude CodeCodex
Design a repository-native change before implementation. Use for architecture, migrations, public contracts, security-sensitive boundaries, cross-cutting work, or decisions with meaningful compatibility risk.
Skill Claude CodeCodex
Run isolated competing attempts at the same brief, pick a base, graft the strongest parts, and verify the synthesis. Use for /patpat-arena, competing designs, or when one attempt would lock the wrong shape.
Skill Claude CodeCodex
Design or scaffold a safe external automation for a concrete provider, trigger, and verification surface. Use when an integration is named and write authority, idempotency, compensation, secrets, and kill-switch behavior must be settled.
Skill Claude CodeCodex
Implement one bounded repository change with a small diff and explicit proof. Use for clear features, localized refactors, and requested fixes after the root cause or design is known.
Skill Claude CodeCodex
Diagnose and fix a reproducible defect through root-cause evidence. Use for broken behavior, regressions, failing runtime paths, and unexplained mismatches when implementation is requested.
Skill Claude CodeCodex
Execute one bounded implementation slice delegated by a named integration owner. Use only when ownership, forbidden scope, allowed mutations, and a proof contract are explicit; do not use for broad direct user requests.
Skill Claude CodeCodex
Evaluate whether an agent skill triggers and performs correctly using isolated, evidence-based trials. Use after non-trivial skill changes or when routing quality is uncertain.
Skill Claude CodeCodex
Trace and prove the blast radius of an existing or proposed repository change. Use for downstream-regression questions, risky diffs, contract consumers, or claims that a small change is safe; remain read-only.
Skill Claude CodeCodex
Investigate a repository and explain how behavior works or why it reached its current shape, including placement, ownership, layering, and critique after an explanation, without making changes. Use for how-does-X-work, where-should-this-live, who-owns-this, are-we-sure, and why-was-this-shaped questions. Do not use…
Skill Claude CodeCodex
Convert a recurring failure, correction, or verification gap from the active task into the smallest durable constraint. Use when the same engineering mistake could recur and should be prevented structurally.
Skill Claude CodeCodex
Run the explicit Patpat evidence-driven engineering loop. Use only when the user invokes /patpat, /patpat-loop, $patpat, or $patpat-loop, or directly asks to activate Patpat mode.
Skill Claude CodeCodex
Diagnose a resource regression or drive sustained numeric optimization against a measurable target. Use for latency, CPU, memory, throughput, startup, or other performance implementation when comparable before-and-after measurement is practical.
Skill Claude CodeCodex
Design a falsifiable workflow for multi-phase repository work or a task with no safe narrow route. Use when several contracts or verification surfaces must be sequenced; remain read-only.
Skill Claude CodeCodex
Independently challenge an implementation, diff, or verification claim without editing. Use after meaningful changes, for explicit code review, or when high-risk evidence needs a separate skeptical pass.
Skill Claude CodeCodex
Drive, checkpoint, validate, resume, or safely take over multi-phase engineering work through a deterministic local state machine. Use for long work, arbitrary branch or handoff pickup, and tasks that must survive context loss without granting delivery authority.
Skill Claude CodeCodex
Inspect the active agent host and install or validate Patpat through its supported native plugin manager or an explicit Agent Skills directory. Use for Patpat installation, discovery, update, removal, or host compatibility questions.
Skill Claude CodeCodex
Commit and open or update a ready pull request after verified work when delivery intent exists, and merge only with explicit land or merge language. Use at the end of an active Patpat implementation and for explicit ship, land, merge, publish, or deploy requests.
Skill Claude CodeCodex
Create or revise a reusable agent skill with precise triggering, progressive disclosure, and behavioral proof. Use when adding a SKILL.md workflow or changing a skill's routing contract.
Skill Claude CodeCodex
Fan out isolated workers across slices or races and return one aggregated report. Use for /patpat-swarm, coverage matrices, parallel checks, or races with a declared selection rule.
Skill Claude CodeCodex
Create or maintain a project-specific verification skill that launches, exercises, observes, and cleans up the real system. Use when a repository lacks repeatable proof of its user-facing behavior.
Skill Claude CodeCodex
Prove an implementation or repository claim against its authoritative artifact or user surface. Use after code changes, during acceptance checks, or whenever tests and builds are insufficient evidence of real behavior.
Skill Claude CodeCodex
Explicit alias for Patpat Loop. Use only when the user invokes /patpat, $patpat, /patpat-loop, or $patpat-loop, or directly asks to activate Patpat mode.
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: