Skip to content

AGENTS.md

Latest commit

 

History

History
115 lines (79 loc) · 5.95 KB

AGENTS.md

File metadata and controls

115 lines (79 loc) · 5.95 KB

Project Conventions for AI Agents

Working Agreement

1. Think Before Coding

Don't assume. Don't hide confusion. Surface tradeoffs.

Before implementing:

  • State your assumptions explicitly. If uncertain, ask.
  • If multiple interpretations exist, present them - don't pick silently.
  • If a simpler approach exists, say so. Push back when warranted.
  • If something is unclear, stop. Name what's confusing. Ask. This applies even when the confusion seems mild ("a bit confusing") or when you can imagine a reasonable resolution. Ambiguity that the agent silently resolves is a class of bug; ambiguity that the user resolves is not.

2. Simplicity First

Minimum code that solves the problem. Nothing speculative.

  • No features beyond what was asked.
  • No abstractions for single-use code.
  • No "flexibility" or "configurability" that wasn't requested.
  • No error handling for impossible scenarios.
  • If you write 200 lines and it could be 50, rewrite it.

Ask yourself: "Would a senior engineer say this is overcomplicated?" If yes, simplify.

3. Surgical Changes

Touch only what you must. Clean up only your own mess.

When editing existing code:

  • Don't "improve" adjacent code, comments, or formatting.
  • Don't refactor things that aren't broken.
  • Match existing style, even if you'd do it differently.
  • If you notice unrelated dead code, mention it - don't delete it.

When your changes create orphans:

  • Remove imports/variables/functions that YOUR changes made unused.
  • Don't remove pre-existing dead code unless asked.

The test: Every changed line should trace directly to the user's request.

4. Goal-Driven Execution

Define success criteria. Loop until verified.

Transform tasks into verifiable goals:

  • "Add validation" → "Write tests for invalid inputs, then make them pass"
  • "Fix the bug" → "Write a test that reproduces it, then make it pass"
  • "Refactor X" → "Ensure tests pass before and after"

For multi-step tasks, state a brief plan:

1. [Step] → verify: [check]
2. [Step] → verify: [check]
3. [Step] → verify: [check]

Strong success criteria let you loop independently. Weak criteria ("make it work") require constant clarification.

Guidelines

Commands

Read the Makefile to discover available targets before running any Go toolchain commands directly.

Gotchas

  • Go is managed by asdf. The go binary resolves through ~/.asdf/shims/go.
  • Architecture doc is the spec. docs/architecture.md (~3600 lines) defines every entity, state machine, algorithm, and validation rule. Read the relevant section before implementing anything. Drift from the spec is a bug.
  • Symphony is prior art, not a template. Sortie derives from OpenAI Symphony but diverges intentionally (Go instead of Elixir, SQLite persistence, adapter interfaces). Do not copy Symphony patterns or Elixir idioms.
  • Workspace safety invariants are security boundaries. Path containment under workspace root, sanitized workspace keys ([A-Za-z0-9._-] only), and cwd validation before agent launch are mandatory — not suggestions. See architecture Section 9.6.
  • Generic naming in core code. Use agent_*, tracker_*, session_* in orchestrator core. Never jira_*, claude_*, codex_* outside their adapter packages.
  • Integration tests are env-gated. SORTIE_JIRA_TEST=1 for Jira, SORTIE_GITHUB_TEST=1 for GitHub adapter integration tests, SORTIE_GITHUB_E2E=1 for GitHub E2E orchestrator tests (also requires SORTIE_GITHUB_TOKEN and SORTIE_GITHUB_PROJECT), SORTIE_CLAUDE_TEST=1 for Claude Code, SORTIE_COPILOT_TEST=1 for Copilot. Without these vars, integration tests must skip cleanly — never fail.
  • SQLite library is modernc.org/sqlite only. Never mattn/go-sqlite3 — CGo breaks the single-binary zero-dependency deployment model.

Boundaries

Always

  • Read the relevant architecture doc section before implementing a feature.
  • Implement adapter integrations as new packages behind the existing Go interface — additive only.
  • Produce a statically-linked single binary with zero runtime dependencies.

Ask first

  • Any change to docs/architecture.md or docs/decisions/*.md.
  • Adding dependencies beyond what the architecture specifies.

Never

  • Modify accepted ADRs in docs/decisions/ without explicit instruction.
  • Use CGo or any library requiring a C toolchain.
  • Put integration-specific logic (Jira field names, Claude Code CLI flags) in orchestrator core packages.
  • Weaken workspace path containment or sanitization rules.
  • Edit LICENSE or README.md without explicit instruction.
  • Do not reference docs/architecture.md, docs/decisions/, section numbers, ADR numbers, or ticket IDs in any comment — godoc or inline. Those belong in specs and plans, not in source files.
  • NEVER prefix commands with GOPATH=..., GOMODCACHE=..., or any Go environment overrides. The asdf shim configures everything.
  • NEVER use /usr/local/go/bin/go, /usr/bin/go, or any absolute path to a Go binary.
  • NEVER downgrade the go directive in go.mod. NEVER add or modify toolchain directives in go.mod unless explicitly asked.

Reference docs

Consult these on demand for the area you are working on, not as a blanket prerequisite to read upfront:

  • docs/architecture-digest.md - a 2-page map of the system for AI agents. Read this document as your first reference during specification, planning, and review. Open the full docs/architecture.md only when the feature you are working on touches one of the areas flagged in the "deep-read" section at the bottom.
  • docs/architecture.md - the full specification (~3600 lines). Read the section that covers the entity, state machine, algorithm, or contract you are about to change. Do not read the document end-to-end before starting work.
  • docs/decisions/ - accepted ADRs. Read when discussing or revising a prior design choice.
  • docs/workflow-reference.md - WORKFLOW.md syntax reference.

Last updated: 2026-04-26

Maintained by: AI Agents under human supervision