Skip to content

AGENTS.md

Latest commit

 

History

History
63 lines (46 loc) · 3.04 KB

AGENTS.md

File metadata and controls

63 lines (46 loc) · 3.04 KB
Raw

IntentCall — Agent Map

IntentCall lets Dart/Flutter apps define agent-callable actions once in AgentRegistry, then project them to MCP, WebMCP, shortcuts, and deep links — without rewriting per transport. Building a Flutter app? Start with mcp_flutter. Building adapters or platform projection? You're in the right repo.

Published docs: docs.page/Arenukvern/intentcall · Full router: docs/start_here/docs_map.mdx

Install Skill Steward meta-skills for this repo:

npx skills add arenukvern/skill_steward

Documentation router

Question Go to
Where is the full doc map? docs/start_here/docs_map.mdx
What does this repo own? docs/NORTH_STAR.mdx
Why is it built this way? docs/DESIGN_FAQ.mdx
How do I use / extend it? docs/DX_FAQ.mdx
Why was X decided? docs/decisions/README.md
How do I contribute? CONTRIBUTING.md
How do I publish to pub.dev? PUBLISHING.md
What is the pre-release status? PRE_RELEASE.md
What skills are installed? .agents/skills/

Non-negotiables

  • Start agent work with steward doctor --json, steward actions list --json, and steward action inspect <id> --json before running declared actions.
  • Use steward benchmark --scenario intentcall.adapter-contract --json for the first Steward dogfood scenario.
  • Do not use legacy steward mcp pipeline execution or steward_run_pipeline_* tools for v1 contracts.
  • Significant design forks → ADR in docs/decisions/ before coding.
  • No secrets, tokens, or private URLs in ADRs or docs.
  • Plan files are temporary — extract durable knowledge to ADR/FAQ then delete.
  • Adapter authors: read intentcall_mcp as the reference implementation first.

Install paths

Agent / Tool Skills location
Antigravity / Claude Code .agents/skills/
Cursor .cursor/skills/ (if using Cursor)
Codex .agents/skills/

Skill authoring detail → .agents/skills/create-skill/SKILL.md

Governance & Skill Steward

This repository strictly adheres to the Cascading Agent Surface architecture governed by Skill Steward. When writing code, documentation, or planning features:

  1. Run steward doctor --json to inspect the v1 contract without executing repository actions.
  2. Run steward actions list --json and inspect intended actions with steward action inspect <id> --json.
  3. Run steward probe --json --profile quick for the safe first pass.
  4. Run steward benchmark --scenario intentcall.adapter-contract --json for the first dogfood loop.
  5. The repository uses standardized agent skills under .agents/skills/ and distributable skills under skills/; skills remain installed separately from hook/plugin wiring.