Imported from satya-anguluri/agent-framework (
AGENTS.md). Install upstream withnpx skills add satya-anguluri/agent-framework. Copyright stays with the author.
Agent Framework Instructions
Purpose
Build a reusable cross-repository context layer for any software project. Repository names, issue prefixes, languages, transports, and database technologies must not be hard-coded into the core.
Rules
- Live repositories are authoritative; the index is only an evidence locator.
- Every claim carries repository, path, and revision provenance.
- Refuse stale evidence and dirty-worktree indexing.
- Never index secrets, generated output, credentials, runtime configuration values, pipeline shell bodies, or tokens. Deployment extractors may retain structural identifiers and references only.
- Evidence is not diagnosis; observed facts stay separate from inferred impact.
- No automatic production deployment.
Extensibility
- Language/framework extraction belongs behind
SourceArtifactExtractor. - Built-ins are adapters, not core assumptions.
- Discover additional extractors with Java
ServiceLoader. - Project topology comes only from configuration and deterministic evidence.
- Work items use a tracker-neutral contract; Jira, Linear, GitHub Issues, and internal systems are adapters.