devarfeen/agent-skills-kit
7 skills7 installsGitHub
Install
npx skills add https://github.com/devarfeen/agent-skills-kitSkills in this repo
1Agents Mdagents-md is the reference slice of the Agent Skills Kit for solo builders standardizing on Antigravity CLI. It is not a feature generator; it is the contract that tells your agent how familiar Claude-style tool names map to Antigravity primitives like read_file, write_file, replace, run_command, grep_search, glob, write_todos, search_web, read_url_content, and subagent dispatch. That matters when you are tired of agents hallucinating tool names or when you migrate a workflow from another agent stack into a workspace-first CLI. The doc also clarifies orchestration: @ references files, not agents, and parallelism comes from the Agent Manager and start_subagent rather than ad-hoc bash loops. For Prism’s audience—one person shipping with agents—the payoff is a single AGENTS.md you can regenerate or extend so every session starts with the same executable playbook. Use it whenever you adopt this kit, refresh workspace bindings, or onboard a new repo to Antigravity without duplicating vendor-specific shims.1installs2Commit Push Closecommit-push-close packages a disciplined GitHub shipping loop for agent-driven fixes and enhancements: validate issue labels, produce a commit message and body that respect human-in-the-loop notes, push, and close the linked issue with an explanatory comment. Solo builders running Claude Code or Cursor against a labeled issue backlog use it when the iteration should end on the issue tracker rather than a pull request—sibling skill commit-push-pr covers the PR plus Closes #N path. The shared policy file is byte-identical across both skills so installs stay self-contained. Label gating is strict: one category label and a ready state, otherwise triage. That makes the skill a Ship-phase launch primitive for small teams that treat issues as the contract between human triage and autonomous implementation.1installs3Commit Push Prcommit-push-pr is a procedural agent skill for solo and indie builders who track work in GitHub issues and want a consistent ship ritual: validate labels on the linked issue, commit and push one iteration, then open a PR whose title and body close the issue (`Closes #N`). It shares byte-identical policy with commit-push-close so you pick PR-based review vs closing the issue with a comment without relearning rules. The skill emphasizes stopping when labels are missing, conflicting, or triage-blocked and routing through `/triage` instead of guessing. It keeps human-decision notes in the commit body or ship output when the issue is `ready-for-human`. Ideal when your agent workflow is issue-driven and you need PRs that auditors and teammates can trace back to a single ready issue.1installs4Feature DiscoveryFeature Discovery is a procedural skill for solo builders maintaining multi-package or multi-service repos who need faithful explanations of how something already works. You invoke it with affected project codes and a concrete topic—an API, config flag, workflow, or bug-shaped behavior—and the agent maps ownership, call paths, and likely rationale while staying out of the editor. That separation matters: discovery output stays in chat, and any CONTEXT.md or artifact change requires explicit approval with quoted proposed text. The skill is optimized for the awkward middle phase when tickets say “fix billing webhooks” but nobody remembers which service owns idempotency keys. It front-loads domain vocabulary so later planning, migration, or refactor skills do not hallucinate module names. Because it forbids side effects, it is safe to run on production branches when you only need orientation. Pair it with implementation or debugging skills after the report looks right.1installs5Feature PromptFeature Prompt is an agent skill for solo and indie builders who have a feature idea, change request, or messy requirement but need a disciplined handoff instead of jumping straight into coding or a giant spec. It compresses the request into a drop-in markdown prompt shaped for grill-with-docs: project context, what is needed, why it matters, observable end result, and only when useful known limits and high-leverage open questions. The skill deliberately stops short of implementation planning; grill-with-docs is where the plan gets grilled, domain language gets aligned with the repo, CONTEXT.md gets updated after approval, and ADRs appear only for hard decisions. When lightweight exploration shows domain terms missing or stale in CONTEXT.md, those candidates are surfaced for user sign-off before any context edit. Best used when you want agent-assisted discovery and critique without prematurely locking architecture.1installs6Orchestrate HerdrOrchestrate-herdr is a tested orchestrator agent skill for developers already running inside the Herdr environment with HERDR_ENV=1. You supply a PRD or parent issue URL and the coding CLI to launch in each worker—examples include codex or claude—and the skill executes a fixed prompt that discovers open sub-issues and spins up one Herdr-managed browser tab per issue. The orchestrator tab stays open, avoids implementing code itself, and follows strict constraints: existing Herdr workspace only, no split panes, no nested agents, and no rewriting the prompt for convenience. Workers run until tests or completion criteria defined in the workflow are satisfied while the orchestrator monitors progress. This fits solo builders and small teams treating a PRD as parallelizable tickets without manually opening and tracking many agent sessions. It is intermediate complexity because you must understand PRD structure, Herdr sessions, and CLI choice. It is narrowly scoped to Herdr; outside that runtime the skill does not apply. Success means every open sub-issue has a dedicated worker with a consistent CLI and observable completion rather than a single monolithic agent chat trying to do everythin1installs7Release Notesrelease-notes is a template-style agent skill that helps solo builders and tiny teams document what actually shipped in language stakeholders and PMs can scan in seconds. It separates a terse Stakeholder Summary—grouped by project code with combined what-changed-and-why bullets—from Detailed Release Notes where each improvement gets Summary, Problem, Change, Impact, Scope, Manual QA Steps, and Commits Included. The format deliberately avoids developer-only framing: changes describe user-visible surfaces, problems state symptoms teams felt, and impact states what is better now. Manual QA gives repeatable verification steps so launch is not only announced but checkable. Use it at the end of a build or fix session when you need consistent release communication across multiple repos or product codes without rewriting the same outline every sprint.1installs