Skip to content

AGENTS.md

Latest commit

 

History

History
80 lines (57 loc) · 3.08 KB

AGENTS.md

File metadata and controls

80 lines (57 loc) · 3.08 KB
Raw

AGENTS.md — Universal Agent Context

This file is the shared entry point for all AI agents operating in the TTA repository. Read this before acting.

What This Repo Is

Therapeutic Text Adventure (TTA) — An AI-powered narrative game for mental health support. Players make choices that branch the story; a therapist/editorial layer curates which paths are safe and effective.

TTA.dev (/home/thein/repos/TTA.dev) provides reusable platform primitives.

Agent Roster

Agent Role Config
Claude Code Main agent — primary decision maker CLAUDE.md
GitHub Copilot IDE suggestions .github/copilot-instructions.md
Augment Code intelligence .augment/instructions.md
Cline AI coding assistant .cline/instructions.md
Gemini Secondary review GEMINI.md

Memory Access (MANDATORY)

Before any non-trivial task, recall context from Hindsight:

  1. mcp__hindsight__recall — query adam-global for durable directives and cross-project patterns
  2. mcp__hindsight__recall — query the current derived project-* or workspace-* bank for TTA-specific context

After completing a significant task (architecture decision, bug root-cause, resolved TODO):

  • Cross-project patterns → retain to adam-global
  • TTA-specific decisions, commands, failures → retain to current derived project-*/workspace-* bank

Hindsight MCP: http://localhost:8888/mcp/ | Start: uvx --from hindsight-api hindsight-local-mcp

Non-Negotiable Standards

  • Python: 3.12+ only
  • Package manager: uv — never pip install
  • Type hints: str | None not Optional[str]
  • Line length: 88 chars (Ruff)
  • Type checking: Pyright standard mode
  • Commits: Conventional Commits (feat:, fix:, refactor:, test:, chore:)
  • Tests: pytest asyncio_mode = "auto"; integration tests use @pytest.mark.integration

Quality Gate (run before committing)

uv run ruff check src/ --fix
uv run ruff format src/
uv run pyright src/
make test

Continuous Testing

make watch        # Reruns unit tests on every .py/.toml save (cancels in-progress run)
make watch-all    # Same, but includes integration tests (requires Neo4j + Redis)

L0 / TTA.dev Interaction Rule

TTA.dev owns the emerging developer-agent L0 control plane.

TTA does not own that layer. In this repo, L0-related work should focus on integrating with TTA.dev, not recreating a second developer control plane inside src/agent_orchestration/.

What TTA agents should do next for L0

  1. keep runtime/player-session orchestration separate from developer-agent orchestration
  2. define how TTA build/test/release work should be represented as TTA.dev L0 tasks and runs
  3. add adapters or scripts that let TTA.dev-managed agents safely operate on this repo
  4. surface enough project/workflow metadata that TTA.dev can coordinate work without touching therapeutic runtime state

Anti-goal

Do not treat IPA/WBA/NGA runtime coordination as the same problem as Claude/Copilot developer coordination. Those are different control planes.