Agent Skills

DocSlime ships agent skills so Codex, Claude Code, ChatGPT, and other skill-aware tools can drive the documentation lifecycle with the same product rules the CLI scaffolds.

The CLI creates files. The skills do the judgment-heavy work: check installation, scaffold a repo, interview the user, fill documents, create ADRs, and review the result for KISS problems.

Install

npx skills add DecisionNerd/DocSlime

For a specific coding agent:

npx skills add DecisionNerd/DocSlime --agent codex
npx skills add DecisionNerd/DocSlime --agent claude-code

Skills

Skill What it does
docslime-install Verifies the docslime binary and installs it if missing.
docslime-init Confirms the current repo and scaffolds the standard docs/ tree.
docslime-fill Interviews the user and fills one scaffolded document.
docslime-adr Creates and fills the next-numbered Architecture Decision Record.
docslime-kiss Reviews filled docs for bloat, stale contradictions, weak traceability, and generic AI prose.

Skill Quality Bar

DocSlime skills are intentionally closer to small operating procedures than prompt snippets. Each bundled skill should tell an agent:

  • When to use it and when to hand off to another DocSlime skill.
  • Prerequisites such as an installed docslime binary, a repo root, or an existing docs tree.
  • Steps that preserve user work and ask for missing facts instead of inventing them.
  • Guardrails for the command surface: init, add, and list are CLI commands; docslime-kiss is a skill.
  • Verification commands or checks that prove the work finished.
  • Failure handling for missing binaries, missing docs, ambiguous targets, dirty context, or unfinished LLM: guidance.

How it works at runtime

Once installed, invoke the skills from inside the agent:

/docslime-install
/docslime-init
/docslime-fill PRODUCT.md
/docslime-adr embed-templates-in-binary
/docslime-kiss

Each skill follows the docs tree instead of maintaining a separate source of truth. Product context lives in docs/PRODUCT.md, design context lives in docs/DESIGN.md, and decisions live in docs/engineering/adrs/.

See the lifecycle guide for how skills fit into continuous discovery, Domain Driven Design, and TDD+BDD traceability.

The same files feed design work: impeccable loads docs/PRODUCT.md and docs/DESIGN.md for homepage/product critique, so skill-driven doc updates improve future UI iteration too.

  1. Run docslime init.
  2. Fill PRODUCT.md with docslime-fill.
  3. Capture discovery in experience/, translate it into REQUIREMENTS.md, then work through architecture, testing, publishing, and observability.
  4. Use docslime-adr whenever a durable technical or product decision is made.
  5. Run docslime-kiss before merging documentation-heavy changes and whenever production learning exposes stale assumptions.

Validation

Run the repo-native checks before release or after changing skill metadata:

cargo test

The repository tests inspect the bundled skills for required sections, so skill robustness stays part of the normal test loop. CI also has a dedicated Agent skills job that checks every bundled docslime-* skill for frontmatter, OpenAI metadata, guardrails, verification, and failure handling. When working in a harness that ships the Agent Skills validator, also run quick_validate.py .agents/skills/<skill> for each changed skill.

Previous Requirements Next Engineering