3.9 KiB
CLAUDE.md — ResolutionFlow
You are Claude Code, the primary coding agent for ResolutionFlow. OpenAI Codex is the resume agent when you hit session or weekly limits.
The first thing to do every session: read .ai/PROJECT_CONTEXT.md, .ai/CURRENT_TASK.md, and .ai/HANDOFF.md. The ritual is spelled out below.
The protocol section below is byte-identical to the shared block in AGENTS.md. If you edit one, edit the other.
Shared protocol
Startup ritual (every session)
- Read
.ai/PROJECT_CONTEXT.md— architectural truth for this repo. - Read
.ai/CURRENT_TASK.md— what we're actively working on. - Read
.ai/HANDOFF.md— exact resume point. - Skim
.ai/DECISIONS.mdfor recent entries relevant to the current task. - Run
git log --oneline -15andgit status. - Before taking action, state back in two sentences: the current goal and your proposed next action.
Handoff ritual (session end — limit hit, task complete, or user stop)
- Update
.ai/HANDOFF.mdto reflect new state. Keep it under ~2K tokens. - If
CURRENT_TASK.mdstatus changed, update it. - If you made an architectural decision, append to
.ai/DECISIONS.md. - Append a session entry to
.ai/SESSION_LOG.md. - If working tree is dirty, commit WIP with
wip(handoff): <one-line summary>. Do not push unless explicitly asked.
Writing rules for .ai/ files
- Use model-neutral voice in
HANDOFF.md,SESSION_LOG.md,DECISIONS.md("previous session did X", NOT "Claude did X" or "Codex did X"). Exception:SESSION_LOG.mdentries include an<agent>field in the header. - Do not duplicate content between files.
CURRENT_TASK.mdholds the goal,HANDOFF.mdholds the resume point,TODO.mdholds the backlog. If unsure where something goes, check.ai/README.md. - Don't invent facts about the repo. If you're uncertain, write
TODO: confirmand flag it.
Project principle
Prefer correct architecture over minimal diff. Flag "simpler approach" tradeoffs for review before taking them.
Claude-specific tooling
GitNexus code intelligence
Indexed as resolutionflow. Earns its cost on cross-cutting work only.
| Tool | When |
|---|---|
gitnexus_query({query}) |
Find code by concept when you don't know where to look |
gitnexus_context({name}) |
Callers/callees of a symbol before touching it |
gitnexus_impact({target, direction}) |
Blast radius before editing shared symbols |
gitnexus_rename({symbol_name, new_name, dry_run: true}) |
Safe multi-file rename |
Use for: core shared symbols (flowpilot_engine, unified_chat_service, auth middleware, get_db, shared hooks), cross-file renames, unfamiliar bug traces, refactor safety. Skip for: new endpoints, isolated fixes, changes you can read in one file.
Re-indexes automatically on commit (PostToolUse hook). Manual refresh if stale: npx gitnexus analyze.
gstack skills
Always use /browse for web, never mcp__claude-in-chrome__*.
Available commands:
- Planning & review:
/autoplan,/plan-eng-review,/plan-design-review,/plan-ceo-review,/plan-devex-review,/devex-review,/review,/cso,/office-hours - Design:
/design-consultation,/design-shotgun,/design-html,/design-review - Browser & QA:
/browse,/connect-chrome,/qa,/qa-only,/setup-browser-cookies - Ship & deploy:
/ship,/land-and-deploy,/canary,/benchmark,/setup-deploy,/document-release - Debug & investigate:
/investigate,/careful,/freeze,/guard,/unfreeze - Other:
/codex(OpenAI second opinion),/setup-gbrain,/retro,/learn,/gstack-upgrade
Git trailer
Every commit: Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Model aliases
Always use alias form (claude-sonnet-4-6, claude-opus-4-6, etc.) via settings.get_model_for_action(). Never hardcode a dated model ID.