CLAUDE.mdandAGENTS.mdare autogenerated by/Init. Do not edit directly -- run/Updateto regenerate.
Developer discipline skills and agents for the forge ecosystem — code quality, testing, defensive programming, debugging, git workflows, and code review.
make install # deploy skills + agents (Claude, Gemini, Codex, OpenCode) and activate git hooks
make validate # run the pre-commit validation chain
make clean # remove build artifactsRequires the forge CLI on PATH (forge-cli). Install via cargo install --locked forge-cli or run INSTALL.md.
| Skill | What it does |
|---|---|
| CodeCleanup | Proactive code cleanup — duplication removal, pattern adoption, readability maximization |
| DataExplorer | Static HTML data explorer apps from SQL or datasets (Vega-Lite, Chart.js, Tabulator) |
| DataVisualization | Choose and configure visualization tools for data explorer apps |
| DefensiveProgramming | Input validation, error boundaries, fail-safe defaults, invariant checking |
| FinishBranch | Decide how to integrate completed work once tests pass — merge, PR, or cleanup |
| FixCI | Inspect failing GitHub Actions checks, pull logs, and propose a fix plan |
| FixIssue | End-to-end issue fixing — triage, diagnose, branch, PR/MR, close (GitHub + GitLab) |
| FixTests | Run tests and systematically fix failing ones using smart error grouping |
| GitHubCLI | GitHub CLI reference — repos, releases, Actions/CI, PR ops, branch protection |
| ReceiveReview | Process code review feedback with technical rigor — verify before implementing |
| RequestReview | Author a scoped review request + CodeReview companion with severity taxonomy and reviewer checklist |
| RustDevelopment | Rust conventions — crate structure, error handling, CLI patterns, config loading |
| SubagentDrivenDevelopment | Delegate independent plan tasks to subagents |
| SystematicDebug | Four-phase debugging methodology — root cause before fixes |
| TestDrivenDevelopment | Red-Green-Refactor, test categories, coverage strategy, anti-patterns |
Shell scripting conventions (BashConventions) live in forge-core alongside MarkdownConventions.
| Agent | What it does |
|---|---|
| ArchitectReviewer | Architecture reviewer — system designs, scalability, integration, evolution |
| CodeReviewer | Senior code reviewer — correctness, security, performance, maintainability |
| ErrorDetective | Error and log forensic analyst — cascade mapping, root cause in distributed systems |
| RefactoringSpecialist | Safe stepwise refactoring with characterization tests and catalog-based transforms |
skills/ skill directories — SKILL.md + optional @-included companions
FixCI/scripts/ optional executable helpers with their own provenance sidecars
agents/ agent markdown files (flat directory)
docs/decisions/ ADRs (structured-madr frontmatter)
rules/ always-loaded behavioral rules
defaults.yaml module defaults
config.yaml user overrides (gitignored)
module.yaml module metadata
.claude-plugin/ Claude Code plugin discovery
.provenance/ SLSA sidecars for every versioned file
Platform directories (.claude/, .gemini/, .codex/, .opencode/) are generated by make install and gitignored. Source of truth is skills/, agents/, rules/, and the adjacent metadata.
- Skills are guidance documents, not tool wrappers. They teach patterns and workflows.
- CodeCleanup is the tactical workflow; DefensiveProgramming provides the principles it references.
- RustDevelopment is Rust-specific; CodeCleanup and DefensiveProgramming are language-agnostic.
- Adopted upstream content ships with a SLSA provenance sidecar under
.provenance/; seeSECURITY.mdfor the audit path and pin-bump policy.
Conventional Commits: type: description. Lowercase, no trailing period, no scope.
Types: feat, fix, docs, chore, refactor, test.
Pre-commit hook (.githooks/pre-commit) runs validation via prek or forge validate with a hash-pinned validate.sh fallback.