- FreePascal only. FPC 3.2.2, Delphi mode, flags centralised in source/units/Shared.inc. Do not introduce another compiled language or repeat compiler directives per unit. - Zero third-party runtime dependencies. The library is RTL + fpjson only. Packages that ship inside FPC 3.2.2 are not third-party: fcl-web is admitted on the same footing as fcl-json (fpjson), and stays confined to MCP.Transport.HTTP. lwpt's testing package is dev-time; anything beyond that needs explicit maintainer approval with a recorded justification. - lwpt is the only…
Audits the current repository for systemic correctness, architecture, churn, simplification, clarity, test value, and operational risks using current source and reproducible probes. It can delegate evidence gathering across bounded capability and perspective lanes. Use when the user runs /codebase-audit or asks for an evidence-backed repository or subsystem audit.
Investigates and creates a project-aligned GitHub issue from a tagline or short description, using the repository's template, evidence, and labels. Use when the user runs /create-issue or asks to file a GitHub issue.
Commits relevant changes, opens a templated draft pull request, fills gaps against the repository's Definition of Ready, fixes CI, and marks it ready. Use when the user runs /create-pr.
Prepares a changelog-first release and, only when explicitly requested, publishes it through the repository's single established release path. Use when the user asks to prepare, cut, tag, or publish a release, bump the version, or generate release notes.
Build and sharpen a project's domain model. Use when the user wants to pin down domain terminology or a ubiquitous language, record an architectural decision, or when another skill needs to maintain the domain model.
Applies the user's git defaults: branch from the remote default, merge rather than rebase, never amend or force-push, and squash-merge pull requests. Use when branching, syncing, committing, pushing, or merging in the user's repos.
Grill the user relentlessly about a plan, decision, or idea. Use when the user wants to stress-test their thinking, or uses any 'grill' trigger phrases.
Turns an unfiled idea into a confirmed mini-spec, implements and validates the selected approach, reviews it, and opens a draft pull request. Use when the user runs /implement-idea or asks to build something without an existing issue.
Validates and implements a GitHub issue against current repository evidence, runs the project's completion gate, reviews the change, and opens a draft pull request. Use when the user runs /implement-issue with an issue number.
Applies the user's FreePascal toolchain and its build, formatting, hook, and test contracts while leaving project-specific mechanics local. Use when scaffolding or working in a FreePascal project that follows this toolchain.
Applies the user's language-agnostic repository layout, documentation, governance, hook, test, agent-file, and changelog conventions. Use when scaffolding or restructuring a repo, writing AGENTS.md or docs, or laying out folders.
Resolves current pull-request review findings in place, validates and pushes fixes, and can autonomously converge and merge an opted-in pull request. Use when the user runs /review-pr or /review-pr automatic-merge.
Reviews a roadmap from fresh project evidence and produces a verified, throughput-anchored version plan, with execution gated on confirmation. Use when reviewing a roadmap, planning releases, or sequencing a backlog.
Reviews a completed workstream from conversation, repository, and forge evidence, uses grilling to agree improvements to delivery speed, process, and codebase health, then applies selected documentation edits and follow-up ticket actions. Use when ending a substantial workstream or running a project retrospective.
Applies the user's ambient engineering bar: current evidence, complete in-scope solutions, reuse, real validation, right-sized value, and maintainability. Use for planning, implementing, debugging, reviewing, refactoring, architecture, or substantial technical investigation.
Commits relevant changes, merges the remote default when needed, pushes the current pull-request branch, and refreshes stale PR metadata. Use when the user runs /update-pr or asks to update a pull request.
Plain text files in a repository that tell a coding agent how the project works: commands to run, conventions to follow and things to avoid. CLAUDE.md, AGENTS.md, cursor rules and skills are the common kinds.
CLAUDE.md or AGENTS.md?
CLAUDE.md is read by Claude Code. AGENTS.md is an open format that Codex, Cursor and other agents read. Many projects keep one and point the other at it.
What is a skill?
A folder with a SKILL.md that describes one capability, such as filling PDFs or reviewing code. The agent loads it only when the task calls for it.
Can I search my own team's files too?
Your agents already can, over MCP, limited to the files you're allowed to read. Searching them from this page is coming.
frostneyAGENTS.md