Scan codebases for implicit rules and generate draft Cohesive substrate artifacts.
npx skills add https://github.com/marktoda/cohesive --skill init-marktoda
Please help me install this Agent Skill. Skill: init Source: https://github.com/marktoda/cohesive/tree/main/skills/init Command: npx skills add https://github.com/marktoda/cohesive --skill init-marktoda
This Skill solves the problem of encoding implicit, undocumented codebase knowledge (such as invariants, gotchas, and engineering conventions) into structured, enforceable Cohesive substrate for teams and AI agents adopting Cohesive for the first time.
Run the init skill on your codebase to automatically scan for implicit engineering rules and generate a draft Cohesive substrate directory for your review.
High-intent search queries and answers about installing and using this skill.
Extracting implicit engineering rules from an existing codebase involves scanning comments, test names, and code patterns to surface undocumented invariants and gotchas. This Skill automates that scanning to produce a bounded set of draft substrate artifacts for team review.
A Cohesive substrate is a structured representation of codebase invariants and engineering rules needed when adopting Cohesive for the first time. You initialize it to translate implicit code knowledge into enforceable draft artifacts like named invariants and behavior matrices.
You can generate draft invariants and behavior matrices for a legacy codebase by scanning for proto-substrate signals in comments and code patterns. This Skill produces a capped set of up to 20 high-signal draft artifacts with side-by-side vocabulary translations for review.
Codebase initialization for substrate artifacts does create optional skeletal top-level documentation like CLAUDE.md and ARCHITECTURE.md if no equivalent docs exist. It focuses primarily on producing draft named invariants, gotchas, and semantic linter candidates.
The best way to document undocumented codebase conventions for AI agents is to scan for proto-substrate signals and translate them into structured substrate artifacts. This Skill generates draft invariants and behavior matrices with side-by-side vocabulary translations to teach teams the substrate categories.
Limitations of automatic codebase substrate initialization include a bounded output of 20 or fewer draft artifacts requiring manual team review and commit. It only surfaces implicit rules from proto-substrate signals and cannot enforce them until drafts are reviewed and committed.
jreakinCatalog unit tests, symbols, and source files from tests/ and src/ directories.
marktodaInventory codebase specifications, invariants, gotchas, linters, and CI checks.
oxhagolliGenerate per-directory CLAUDE.md files and refresh top-level CLAUDE.md and ARCHITECTURE.md.
marktodaGenerate codebase-grounded design options and pressure-test them with 25 questions.
