Skip to content

Add CLAUDE.md to provide agent context for future tasks #5

Description

@fullsend-ai-retro

What happened

PR #4 (link) was a docs-only change adding CONTRIBUTING.md. The code agent produced good output, but added a "Running the project locally" section (lines 20–25) that was not in the acceptance criteria from issue #3. The review agent flagged this as [scope-creep-minor] at severity low. The repo currently has no CLAUDE.md, AGENTS.md, or .claude/ directory. All agent roles are enabled (triage, coder, review, fix, retro, prioritize) via .fullsend/config.yaml, indicating this repo will see ongoing agent activity.

What could go better

The code agent went slightly beyond the acceptance criteria by adding content the issue didn't request. While the addition was helpful and the review agent correctly flagged it as minor, this pattern could lead to more significant scope creep on complex tasks. Without a CLAUDE.md, agents have no repo-specific guidance on conventions, scope behavior, or development patterns. For this simple docs task the impact was negligible, but as the repo grows and receives more complex issues, agents will lack context about project structure, testing expectations, and coding conventions. Confidence: moderate — this is a proactive improvement rather than a response to a failure. The risk is low and the benefit scales with repo complexity.

Proposed change

Create a CLAUDE.md file at the repository root with sections covering: (1) project overview (Node.js REST API for task management, zero external dependencies, uses built-in node:http, node:test, node:assert), (2) directory structure (src/index.js — HTTP server, src/store.js — in-memory TaskStore, src/store.test.js — unit tests), (3) development commands (npm start, npm test, npm run dev), (4) coding conventions (no external dependencies, use Node.js built-ins only), and (5) scope guidance (stick to acceptance criteria from issues; additions beyond the spec should be noted in the PR description). This gives agents the context they need to produce well-scoped, convention-aligned code on the first attempt.

Validation criteria

After CLAUDE.md is added: (1) the next 3 code agent runs on this repo should not trigger scope-creep findings from the review agent, and (2) code agent PRs should follow the documented conventions (e.g., no external dependency additions, correct test patterns). Verify by checking review agent findings on the next 3 PRs for scope-creep flags.


Generated by retro agent from #4

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions