vignesh-nagarajan-vn
- 9 skills
- 0 followers
- 18 hours ago last updated
- ▌ Docs Sync · vignesh-nagarajan-vn bundleWork out which documentation a code change made stale and update only those files, using a per-repo .claude/docs-map.yml instead of re-reading every doc. Use after changing behavior, routes, commands, environment variables, dependencies, schemas, or UI; when asked to update or check the docs; and before calling any change done in a repo that keeps its docs current.
- ▌ LLM Evals · vignesh-nagarajan-vn bundleBuild and run eval harnesses for features that call a model, with a keyless mock lane in CI, a budget-capped live lane behind a key gate, thresholds in a versioned file, and judge scoring. Use when adding or changing evals, testing a prompt or model swap, gating a release on output quality, or investigating a quality regression. Runs scripts/check_thresholds.py.
- ▌ Repo Scout · vignesh-nagarajan-vn bundleGet oriented in an unfamiliar or large repo from one compact map instead of reading files. Use at the start of work in a repo with no CLAUDE.md, when asked what a codebase does or where something lives, before any broad or cross-cutting change, and whenever the alternative is opening several files just to find the entry point. Runs scripts/repo_scout.py.
- ▌ Claim Trace · vignesh-nagarajan-vn bundleCross-check every number in a paper, report, or README against the run artifacts it should have come from, and flag the ones with no source. Use before submitting or publishing anything containing results, after re-running experiments, when a results directory has been regenerated, and when asked to verify or audit reported figures. Runs scripts/trace_claims.py.
- ▌ Prose Guard · vignesh-nagarajan-vn bundleEnforce this user's writing conventions on any human-facing English, chiefly no em dashes, no AI-tell vocabulary, no hedge stacking, and no bullet bloat. Use when writing or editing a README, docs, CLAUDE.md, commit message, PR body, or release note, and whenever asked to review, tighten, de-slop, or humanize writing. Detection runs through scripts/prose_lint.py rather than by reading files.
- ▌ LLM Pipeline · vignesh-nagarajan-vn bundleArchitecture rules for code that calls an LLM. Code owns the decision, schemas validate every model output, keys stay server-side, and a mock provider keeps demos and tests keyless. Use when adding or changing an LLM call, designing a multi-pass or multi-agent pipeline, choosing which model handles which pass, or reviewing an app that talks to Claude or another provider. Runs scripts/audit_llm_app.py.
- ▌ Readme Brand · vignesh-nagarajan-vn bundleWrite or restructure a README in the house style, with a centered header block, grouped shields.io badge rows, a live demo link, and a screenshot table, keeping the body lean and deferring detail to docs/. Use when creating a README, restyling one, adding or fixing badges, or checking that a markdown file's links, images, and heading anchors still resolve. Runs scripts/check_readme.py.
- ▌ Agent Contract · vignesh-nagarajan-vn bundleWrite, refresh, or audit a repo's CLAUDE.md onboarding contract so future sessions start from it instead of re-exploring. Use when a repo has no CLAUDE.md and will be worked in again, when asked to update or check it, when its paths or commands have gone stale, and after any change to layout, commands, or conventions. Runs scripts/verify_claude_md.py.
- ▌ Session Handoff · vignesh-nagarajan-vn bundleWrite or read a HANDOFF.md so a new session starts from a short state summary instead of re-exploring the repo. Use when wrapping up, when the user says handoff, pause, stop here, or continue later, when context is running out mid-task, and at the start of a session in a repo that already has a HANDOFF.md.