oliver-chase
UserShared governance layer for Claude Code: commit gates, contracts, and skills that load automatically in every repo you work in.
Categories
Indexed Skills (18)
behavior-preserving-module-extraction
Use when the user says a script is 'too big', 'needs to be split up', 'extract a module from this', 'pull this logic into its own file', or wants a large script refactored into smaller pieces without changing behavior. Uses wrapper delegation and dependency injection so existing call sites keep working.
code-security-review
Review code for real security vulnerabilities — injection, auth/authz bypass, insecure deserialization, secret handling, SSRF, path traversal, XSS, insecure direct object references. Use when the user asks for a security review, mentions a vulnerability class by name, or when touching auth, input parsing, file paths, external requests, or serialization.
deep-review
Post-build deep review loop — single sequential review agent over a commit range, simplify pass, validate-then-DELETE legacy cleanup, live re-verification on the real deploy, and a live session tracker so coverage and current state survive the session. Run before closing any multi-commit build session.
diagnosing-bugs
Diagnosis loop for hard bugs and performance regressions. Use when the user says "diagnose"/"debug this", or reports something broken/throwing/failing/slow.
docs-reality-audit
Use when the user asks 'are these docs accurate', 'audit the docs', 'check the README against the code', 'do these docs contradict each other', or reports doc drift/stale instructions. Cross-checks every factual claim against live code, flags doc-vs-doc contradictions, consolidates to one file per concept.
functionality-audit
Full functionality audit of a live product repo — crash-safe incremental execution, quick wins fixed inline, larger findings written as context-free executable specs. Covers dead code, broken flows, doc truth, RLS/security, tooltips, revert symmetry, and marked-complete-but-isn't.
olivercode-adoption
Use when adopting OliverCode into a machine — the user forked or cloned it and says 'set this up', 'onboard my repos', 'point my CLAUDE.md files at OliverCode', 'audit my Claude config', or opens a fresh OliverCode checkout wanting it running. Walks the whole install: discover what exists, classify it, install activation, register the fleet, install enforcement, prove the gates block. Works for a clean machine or a messy existing setup. Optimizes and points; never amputates.
olivercode-cli
Use when modifying the OliverCode CLI launcher: command structure, help output, update reliability, self-locating path resolution, and launch-path validation. Load before any CLI change.
repo-audit-with-execution-batches
Run a structured 7-part repo audit that discovers FIX/PATCH/TODO/HACK/FIXME markers, structural gaps, doc staleness, and code-quality issues.
repo-structural-cleanup
Execute large-scale file-move, file-rename, and file-delete operations without breaking imports, references, or git history.
tdd
Test-driven development. Use when the user wants to build features or fix bugs test-first, mentions "red-green-refactor", or wants integration tests.
validator-consolidation
Consolidate overlapping validate-* scripts into grouped validators without breaking runners, docs, or nested-repo git boundaries.
automation-first-agent-design
Use when designing or reviewing a cron job, scheduled pipeline, or agent workflow, or when the user asks 'should this be automated', 'can we make this deterministic instead of LLM-based', 'reduce LLM calls in this pipeline', or 'add a fail-closed guard'. Prefers deterministic flows and zero-LLM paths over agent calls; enforces fail-closed behavior on any gate.
model-routing
Use when deciding which LLM tier to call for a step (cheap model for classification/validation vs. capable model for generation), when the user asks 'which model should this use' or 'can we use a cheaper model here', or when configuring LLM_PROVIDER_ORDER. Pattern contract, not implementation.
never-guess
When you don't know something, never guess — ask or look at the actual data.
plan
Use when the user says 'make a plan', 'write this up as a plan first', or 'don't start coding yet, just plan'. Writes a structured plan (Goal/Steps/Risks/Definition-of-Done) as a markdown file to <repo>/docs/plans/; does not execute. Distinct from Claude Code's built-in ephemeral Plan Mode, which writes no file.
repo-discovery
Use when the user says 'explore this repo', 'what does this codebase do', 'give me an overview of this repo', 'onboard me to this project', or opens work in an unfamiliar repo before making changes. Read-only: surface scan, entry points (README/CLAUDE.md/ARCHITECTURE.md), source tree, key files.
structured-code-removal
Remove a feature or module from a codebase by deleting the function, all call sites, dependent functions, signature updates, docstrings, validators, and tests.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.