Personal agent skills, written and battle-tested against real daily work
across a multi-agent setup (Hermes, Codex, OpenCode, Command Code, and
more). Each skill is a standard SKILL.md package — installable by any
agent harness that supports the Agent Skills format.
# everything
npx skills add kartikkabadi/skills
# or one at a time
npx skills add kartikkabadi/skills/handle-thisSkills live at the repo root, one directory per skill. You can also just
copy any skill directory into your harness's skills folder
(~/.agents/skills/, ~/.hermes/skills/, .claude/skills/, etc.).
| Skill | What it does |
|---|---|
| handle-this | Drop any link, file, screenshot, or idea on the agent. It identifies what it is, mines your local context (screen history, past agent sessions, repos), and acts on it completely — no steering needed. |
| decide-for-me | Cognitive offload. Sorts the pile with an Eisenhower grid, executes everything safely delegable, and escalates at most 3 yes/no decisions with a recommendation and a safe default. |
| agent-council | Assembles one read-only subagent per skill lens to vote on a contested decision; later waves see earlier findings, then a tally and report. |
| call-advisors | Calls the fixed WATCHDOG advisor roster (Architecture, Security, Fixer, Verifier, and the rest) for review on a plan, decision, or finished work, in a native background loop or an on-demand council. |
| grade-it | The verifier. Finished work gets graded by independent fresh-eyes subagents against a rubric built from the original request — not the author's effort. Nothing reports "done" until it passes. |
| Skill | What it does |
|---|---|
| agent-input-trust | Treats external content (webpages, repos, emails, MCP responses) as evidence, never instructions. Anti-prompt-injection discipline. |
| how-to-code | Loaded before any code writing, editing, planning, or reviewing. |
| clean-code | Practical clean-code discipline for production software. |
| hygiene | Cleanup discipline: close only what you opened, kill only what you started, leave the environment as found. |
| git-branch-worktree-discipline | Branch/worktree safety, destructive-command boundaries, parallel-agent isolation. |
| supply-chain-install-protection | Install-time package security: minimum release age gates, Socket Firewall shims, dependency audits. |
| Skill | What it does |
|---|---|
| link-research | Drop a link (X, YouTube, article) and get grounded research of it. |
| todo-mastery | Phased, heavily decomposed task lists with real-time tracking, from one-session tasks to 1000-todo programs. |
| youtube-transcript-to-markdown | Full YouTube transcript to a polished, chaptered markdown file. yt-dlp captions (whisper fallback), parallel-chunk rewrite subagents, verified output, zero artifacts left. Ships a reusable VTT rolling-caption dedupe script. |
| repo-inspection | Deep codebase audits: structure, dependencies, what's built, partial, missing, blocked. |
| find-skills | Locate, evaluate, and install existing skills before building from scratch. |
| private-intelligence-reader | Build private, source-grounded intelligence readers from feeds and vault notes, with anti-slop UI gates. |
| unit-economics-analysis | SaaS unit economics, pricing, margins, break-even, sensitivity — verified numbers only. |
| telephony-product-architecture | Design and feasibility-check phone-call products: carriers, CPaaS, iOS/Android constraints, AI call handling. |
Every skill is a directory with a SKILL.md (YAML frontmatter with name
and description, then the operating instructions). Some ship helper
scripts under scripts/. Works with any harness that reads the Agent
Skills format — Claude Code, Codex, Cursor, Hermes, OpenCode, and the rest
of the skills.sh ecosystem.
MIT. See LICENSE.