Borda/.home Skills

Analyze GitHub issues, PRs, and repo health for an OSS project. Summarizes long threads, assesses PR readiness, detects duplicates, extracts reproduction steps, and generates repo health stats. Uses gh CLI for GitHub API access. Complements oss-maintainer agent.

Comprehensive config audit for the entire .claude/ directory. Orchestrates self-mentor across all agents, skills, settings, and hooks to detect correctness issues, broken cross-references, interoperability problems, infinite loops, redundancy, and inefficiency. Reports findings by severity and auto-fixes critical, high, and medium findings; CLAUDE.md contradictions and missing permissions are always reported but never auto-fixed.

Calibration testing for agents and skills. Generates synthetic problems with known outcomes (quasi-ground-truth), runs targets against them, and measures recall, precision, and confidence calibration — revealing whether self-reported confidence scores track actual quality.

Delegate narrow, mechanical coding tasks to OpenAI Codex CLI — Claude orchestrates and judges, Codex executes. Pre-flight checks ensure graceful degradation on machines without Codex.

TDD-first feature development orchestrator. Analyses purpose, scope, and codebase compatibility before writing a single line of implementation — starts with a demo use-case doctest/test to nail the API contract, then drives implementation through TDD, and finishes with doc updates, QA, linting, and a full review pass.

Bug-fixing workflow — diagnose the problem, reproduce it with a regression test, apply a targeted fix, then verify with linting, quality checks, and optional optimization.

Create, update, or delete agents and skills with full cross-reference propagation. Also manages settings.json permissions atomically with permissions-guide.md via add/remove perm operations.

Analyzes ongoing work patterns and the existing agent/skill roster to suggest creating new agents or skills for specialized or repetitive tasks. Continuously monitors what tasks are being done repeatedly or where specialist knowledge would help. Avoids recommending duplicates.

Performance deep-dive orchestrator. Establishes a baseline, spawns perf-optimizer agent to identify the real bottleneck, and produces a before/after report. Covers CPU, memory, I/O, concurrency, and ML/GPU workloads.

Test-first refactoring orchestrator. Ensures test coverage exists before changing code — adds characterization tests if missing, then applies logic improvements, API cleanup, and structural changes with verified input/output consistency.

Prepare release communication from git history, PRs, or a diff. Adapts output to context — user-facing release notes, CHANGELOG entry, internal release summary, or migration guide. Groups changes by type, filters noise, writes in plain language for the audience.

Implement applicable PR review suggestions via Codex — fetch GitHub PR review comments, classify each as implement/skip/flag, dispatch actionable ones straight to Codex, then validate with lint, types, and tests.

Multi-agent code review covering architecture, tests, performance, docs, lint, security, and API design.

Security audit of code or a feature. Checks OWASP Top 10, Python-specific vulnerabilities, ML security concerns, authentication/authorization, secrets handling, and dependency vulnerabilities. Flags issues by severity with specific remediation steps.

Survey SOTA literature for an AI/ML topic, method, or architecture. Finds relevant papers, builds a comparison table, and recommends the best implementation strategy for the current codebase. Delegates deep analysis to the ai-researcher agent.

Drift-detect and sync git-tracked .claude/ and .codex/ config from project to home. Default shows what would change; "apply" performs the sync.