odin-codex-plugin

Verbalized Sampling (VS) protocol for intent exploration before planning, mode-aware. Default `exhaustive` runs full VS; `collaborative` runs tip-sharing dialogue; `adversarial` walks the design tree one fork at a time. Auto-detects from phrasing ("help me refine" → collaborative, "poke holes" → adversarial); override via `/askme adversarial|collaborative|exhaustive`. Use for ambiguous tasks or maximum clarifying questions before committing.

Tidy-operations dispatcher. Invoke on "tidy", "clean up", "tidy this file/memory/workspace/git/docs", or when active context (current file, diff, stack, memory directory) has structural rot to resolve before touching behavior. Detects target domain from context and routes to the sibling skill. Requires explicit target or clear active-context signal — do not invoke speculatively.

Read-only codebase exploration: discovery, structural reading, and emission of architecture/pattern/tooling/dependency summaries. Use to understand existing code, map files, trace function flow, locate symbols, or build pre-implementation context. Uses 1/3/5 Explore-agent escalation scaled to scope. Trigger on "explore", "find where X is", "how does X work in the code", "map the codebase", "what files handle Y", or any architecture/pattern/tooling/dependency context request on a local repo — even without naming /explore.

Surface deepening refactors that turn shallow modules into deep ones, informed by `CONTEXT.md` and `docs/adr/`. Use when the user asks to improve architecture, find refactor candidates, raise testability, or make a codebase more agent-navigable. Skip for single localized fixes.

Surface in-task-collaboration protocols when the user describes an AI workflow informally — URL-as-entity-reference, PR-comment threads as session memory. Trigger when the user names entities by colloquial label instead of stable URL, asks "how should I structure this for Claude", or describes a multi-step Claude workflow without a durable handle. Apply reactively, not as a checklist.

Code search, analysis, and refactoring using ast-grep (sg). Use for AST-based code modifications, structural search, and linting.

Verbosity-reduction response register. Drops redundant clauses, narrative hedging, and ceremony while preserving articles, grammar, and decision-oriented register. Trigger on "caveman", "compact mode", "less tokens", "be brief", or context-window pressure in long sessions.

Reduce concepts, duplication, and ceremony in internal code while touching nearby code. Use when working an existing path and you spot dead fields, redundant wrappers, or speculative abstractions; distinct from refactor-break-bw-compat (internal hygiene, not public API removal).

Auto-router for context gathering. Detects whether the task needs codebase exploration or external knowledge research and dispatches accordingly. Trigger on "get context", "background on X", "context on X", "how does X work", architectural orientation, or any setup-before-coding request.

Set visual and interaction direction for any UI surface (web, React, TUI, CLI, desktop, Qt, design-system tokens) before any UI code. Direction-first: generates 3-4 distinct directions via verbalized sampling, picks one via per-axis single-select, then derives palette, typography, spacing, motion budget. Loads when the user asks for UI work, palette/theme/tokens, mentions a design system, or when output looks AI-generic, vibe-coded, sloppy, or default Tailwind/shadcn/Bootstrap. Enforces two-sided anti-slop charter: forbids purple gradients, `transition: all`, system-ui, default Tailwind palette AND overkill compensation (sprites, gradients everywhere, animation on every element).

Polymorphic iterative repair loop — accept a verifier failure, structured findings (review/resolve/triage-issue), or a bug description; modify→verify→keep on green, auto-revert on guard regression, until clean or iteration cap. Use when the user says "fix", "make it pass", or "apply the findings", or hands an artifact + repo and expects patches; auto-routes to gh-fix-ci or gh-address-comments when an open PR + gh auth + GH-flavored input is detected.

Restructure Web-UI / human-triggered tasks into CLI + file-output loops the LLM can iterate alone, with structured logs and addressable scratchpads. Apply trap-or-abandon: if a step cannot be looped, improve the harness rather than babysit. Trigger on iterative grunt-work, "push a button in a web UI to trigger this", monitoring dashboards, or any workflow whose inner loop requires a human in the middle.

Hotspot detection and performance root-cause analysis: flamegraph interpretation, allocation tracking, latency profiling, regression measurement. Use when a workload misses its latency, throughput, or memory budget; when a benchmark regresses; or before optimizing any hot path (no optimization without a profile).

Conversational QA mode — user reports bugs in plain language, agent clarifies minimally, files GitHub issues that survive refactors. Trigger on "QA", "QA session", or ad-hoc bug reporting without a fixed deliverable shape. Distinct from branch-scoped and PR-scoped review.

Plan a refactor as a sequence of tiny, working commits via adversarial interview. Default output is a markdown plan at `<project-root>/docs/refactor-plans/<name>.md`; pass `--emit-issue` to also file a GitHub issue. Trigger on structural refactors (rename, extract, move, split, dedupe) — NOT new features.

Read-only external knowledge gathering via ODIN's 5-tier doc ladder (Official docs → API refs → Books/papers → Tutorials → Community). For library APIs, framework behavior, SDK migrations, version-specific docs, vendor announcements, RFCs. Verifies claims against primary sources. Invoke on "how does X library work", "migration guide", "docs for", or any named library/framework/SDK/API/CLI/service.

Resolve code review comments by verifying validity, then proposing multiple architectural solutions (not naive fixes) for confirmed issues. Use when addressing review feedback or analyzing comment validity.

Adversarial security audit — STRIDE, OWASP Top 10, supply-chain (CVE/SBOM), secrets scan, auth/authz analysis. Use on changes touching auth, input parsing, deserialization, network I/O, dependencies, or secrets; before any production release or external-surface PR.

Install git pre-commit hooks via the project's hook tool — Husky+lint-staged (JS), pre-commit (Python/OCaml), lefthook (Go), cargo-husky (Rust). Use when the user wants commit-time formatting, linting, type-checking, or test gates. Detects ecosystem first.

Build safe, syntax-aware srgn CLI commands for source-code search and transformation. Use for srgn commands, scoped refactors (comments/docstrings/imports/functions), multi-file rewrites with --glob, tree-sitter queries, or CI checks with --fail-any/--fail-none.

One-shot bootstrap of strict-mode tooling per ecosystem plus per-task GOALS.md scaffolding so an agentic loop can self-verify. Writes typechecker/linter/schema-validator config for TS (strict + noUncheckedIndexedAccess + exactOptionalPropertyTypes), Python (Pyright strict, Ruff strict), Rust (Clippy deny-correctness), Go (golangci-lint with staticcheck), OCaml (dune --release); establishes `.agent-tasks/<id>/GOALS.md` per-task convention distinct from project-stable AGENTS.md. C++/Java/Kotlin and framework specifics (Spring Boot, Nest, React-strict) are out of scope. Trigger on new project bootstrap, agentic-task setup, "make this self-verifying", "set the loop's goal", "scaffold goals for this issue". Pairs with `llm-self-loop` runtime.

Cross-domain taste skill — apply distinctive judgment to any artifact (prose, code, design, decisions) instead of converging to AI defaults. Two modes — `audit` (judge work against the two-sided charter and portable anchors) and `anchor` (load register before producing). Auto-detects by phrasing; override via `/taste audit | anchor`. Trigger on "is this slop?", "overkill?", "elegant?", "taste-test this".

Test-Driven Development (TDD) across any supported language. Use when implementing features or fixes with TDD methodology, writing tests before code, or following XP-style development.

Two-party posture — user as director, agent as executor; every fork, tradeoff, or choice surfaced via batched AskUserQuestion with a recommended default. Use when the user invokes /duet, says "ask before" / "pair with me" / "human-in-the-loop", or for aesthetic/architectural/irreversible decisions.

ODIN's compact-form conversation skill -- formal-logic English register with predicate claims, Hoare-triple framing, and ASCII shortened-English keywords. Trigger when user requests "axiom", "axiom-mode", "axiom-compact", or "compact form".

Evidence-first generator for a personal <name>-taste Claude Code skill. Mines local memories and agent histories for influences, slop bans, and overkill bans; asks compact confirmation forks; previews the synthesis; then writes a right-sized taste skill with exactly 5 anchors by default. Trigger with "generate my taste skill", "make my taste skill", or "derive my taste spine".

Help address review/issue comments on the open GitHub PR for the current branch using gh CLI; verify gh auth first and prompt the user to authenticate if not logged in.

Inspect GitHub PR checks with gh, pull failing GitHub Actions logs, summarize the failure, then plan and implement the fix after user approval. Use when the user asks to debug or fix failing PR CI on GitHub Actions; external checks (Buildkite, etc.) are reported as URLs only.

Audit memory directory for structural issues (orphans, dangling refs, duplicates, missing sections, oversized entries) and staleness against session-history transcripts; report-first, fix-on-confirmation. Use when the user says "audit memory", "memory hygiene", or "find stale/duplicate memories".

Produce share-safe copies of memory files under /tmp with PII redacted (paths, emails, session IDs, dates) and credentials scanned (tokens, keys); never mutates originals. Use when the user says "sanitize memory for sharing", "redact memory PII", or "scan memory for credentials".

Scan agent's session-history transcripts for save-worthy signals (corrections, preferences, decisions, references), propose and write auto-memory files with valid frontmatter and MEMORY.md entry. Use when the user says "save this to memory", "remember that", or "scan this session for memories".

Decompose a plan, PRD, or spec into independently-grabbable vertical-slice issues (markdown file by default, GitHub issues via flag). Trigger when the user wants implementation tickets, work decomposition, or to convert an implementation plan into parallelizable work. Takes a plan file and emits atomic vertical slices.

Synthesize current conversation context and codebase understanding into a PRD artifact (markdown file by default, GitHub issue via flag). Trigger when the user asks for a PRD, requirements doc, feature spec, or wants to crystallize an in-flight discussion into a durable artifact before planning. PRD precedes implementation planning.

Investigate a reported bug to root cause, then emit a TDD-shaped fix plan as an issue artifact. Trigger when the user reports a bug, says "triage", asks for issue investigation, or wants a fix plan before code changes.

Extract a domain glossary from the current dialogue; flag ambiguities, propose canonical terms, persist to `UBIQUITOUS_LANGUAGE.md`. Trigger when the user is hardening domain terminology, building a glossary, or fresh domain concepts surface in conversation without documented language.

Grill against the existing domain model. Stress-test a plan's terminology against `CONTEXT.md` and ADRs; update both inline as decisions crystallise. Trigger when user proposes a feature/refactor that touches business concepts and the project has documented domain language to honor — or when domain language is missing and needs capture.

Mechanically tighten existing prose — restructure sections by dependency order, split or merge paragraphs, remove redundancy. Use to compress verbose plan files, READMEs, ADRs, and design docs. Does NOT change voice, register, tone, or any ODIN-mandated phrasing.

Author a single new skill — produce a SKILL.md plus optional bundled references and scripts following Anthropic's progressive-disclosure conventions. Trigger when the user asks to "write a skill", "create a skill", "draft a SKILL.md", or "add a skill" for a specific capability. Distinct from repo onboarding workflows that write AGENTS.md and project conventions.

Hypothesis-driven defect isolation — stack-trace forensics, breakpoint strategy, state inspection, and root-cause confirmation via minimal repro. Use when a defect surfaces (test failure, crash, exception, wrong output, intermittent flake) and the cause is not immediately obvious from the change set.

Proof-driven development. Use when implementing with formal verification using property-based testing, theorem proving, or proof tactics; zero unproven property policy enforced.