Codebase Knowledge Builder 1.0.0

Codebase Knowledge Builder turns an unfamiliar repository into durable, evidence-backed working knowledge for humans and coding agents.

This first production release rebuilds the skill around a strict trust boundary, bounded investigation, source citations, explicit uncertainty, and reusable Markdown artifacts. One canonical Agent Skill now serves Skills CLI, Codex, Claude Code, Hermes Agent, npm, and ClawHub distribution without adding an agent-specific runtime.

Highlights

• Safe, read-only analysis of repository content as untrusted evidence
• Observed, inferred, unknown, and not-applicable evidence labels
• Revision receipts, scope boundaries, exclusions, and path:line citations
• Deterministic reconnaissance, deep-dive, synthesis, and delivery phases
• Output containment that rejects symlinks, junctions, mounts, reparse points, collisions, and unapproved overwrites
• A complete worked example against a fixed source revision
• Content-only npm distribution with no dependencies, lifecycle scripts, executable, or runtime entry point
• Thin Codex and Claude Code manifests plus portable Hermes and Skills CLI routes
• Three-platform GitHub Actions validation on Ubuntu, macOS, and Windows
• Rebuilt README, governance files, security reporting, llms.txt, and original dithered repository artwork

Install

npx skills add OthmanAdi/codebase-knowledge-builder --skill codebase-knowledge-builder

The README documents pinned, npm, Codex, Claude Code, and Hermes installation routes plus exact compatibility evidence and current limitations.

Security

Repository instructions are data, not authority. The skill does not execute target instructions, collect secrets, follow linked paths outside scope, or write into the analyzed repository unless the user explicitly chooses a safe target-repository output.

See the security policy for private reporting and the worked example for the evidence contract in practice.