skills

Home for everything involved in developing agent skills that reproduce @fuj1g0n's development environment and process: the research (docs/research/) and the resulting skills (skills/), which are the deliverables. The repository is an Agent Plugins package with a root plugin.json. Distribution format and consumption management are independent: prefer Agent Plugins for a shared, vendor-neutral structure across compatible tools, and APM for managing dependencies and deployment across explicitly selected tools. Format portability does not guarantee support for every component in every tool.

Not every investigation yields a skill; research snapshots are kept as records either way.

Skills

Tool-management skills are named for scope, environment, and use case (project-*, linux-*, windows-*), not for the package manager behind them. Project development has exactly one source of truth. It is mise unless the project is built and distributed through Nix, in which case Nix also provides the devShell. A Nix devShell without package outputs is a migration candidate, and only Nix-to-mise migration is supported. User-wide tools prefer an official self-upgrading per-user binary, otherwise Nix in Linux or WSL and WinGet user scope on native Windows. Toolchain skills may pin a task runner but do not define recipes. See project-toolchain-selection.

Design workflow

software-design replaces the former adr entry point. Use it for an explicit design question, design document, or decision record; routine coding does not require a separate design phase or new document. ADR remains an output format, with existing collections and templates preserved. Fresh logs do not automatically get a format-adoption record.

When pstack architect already owns design, its rationale and sketch stay authoritative; the skill consumes them without repeating analysis or adding approval. Ponytail's active level continues to constrain scope and simplicity. See the integration adapter for boundaries, including optional execution decision trails.

Installations obtain the replacement when they update their pin to a commit that includes it.

Install

APM

APM is the preferred consumption manager, not the distribution format. Select targets explicitly in the consumer manifest and verify the components each tool receives. On the tested Copilot path, APM installs the repository as a native skills@apm Agent Plugin without transitive dependencies. Declare it in apm.yml (~/.apm/apm.yml for user scope) and run apm install (-g for user scope); APM rejects imperative installs of Agent Plugins from a local path.

dependencies:
  apm:
    - fuj1g0n/skills#<full-40-char-sha>

The apm-usage skill that skill-management defers to is not bundled; declare microsoft/apm/packages/apm-guide#<full-40-char-sha> separately.

Update the dependency's SHA in the maintained manifest source. For project scope, run apm install from that project. For user scope, run apm install -g. Do not install or update skills@apm with copilot plugin commands. If an older setup installed skills@dotfiles, uninstall that plugin before switching to APM.

Pin with a full 40-char commit SHA; short SHAs and floating branches do not resolve reliably.

Native Copilot marketplace

APM is the default recommendation, not a requirement to replace an adopter's explicit manager policy. A native Copilot marketplace can consume this Agent Plugins package directly. An APM failure is not a prerequisite. For a native marketplace, declare fuj1g0n/skills as a GitHub source pinned to a full commit SHA, then use copilot plugin install skills@<marketplace>. Use the Copilot marketplace reference for the entry format. Do not install the same package through two managers in the same tool and scope. APM's generated marketplace remains APM-managed consumption. Record compatibility failures and workarounds in the affected consumer.

Development

The instruction scopes separate reusable adopter workflows from this repository's maintenance procedures and consumer-specific configuration. Read AGENTS.md for contributor instructions.

Development tools and tasks are defined in mise.toml and pinned in mise.lock, including the npm dependency graph under .mise/locks/.

mise trust
mise install
mise run audit   # skill contract audit and routing corpus
mise run lint    # Markdown lint

Evaluation profile

This collection's behavior-evaluation profile is a maintainer choice. It is not a requirement for users of the distributed skills. Run each required scenario with fresh agents on both targets:

Use the shared evaluation method. Report unavailable targets as unverified. Mechanical checks or repeated runs on one model do not satisfy this cross-model profile. Publish only when requested. A release does not itself update any consumer.

Documentation

docs/ is an Open Knowledge Format (OKF) v0.2 bundle.