Project-aware development context tracker for coding agents (Claude Code, Codex). Logs git activity, tracks todos, maintains branch notes, and updates CLAUDE.md and AGENTS.md.
The server defines 3 tools with reasonable naming conventions (verb_noun pattern), but definition quality is inconsistent. All tools have descriptions and input schemas visible in src/index.ts, but several gaps exist: parameter descriptions are sparse or missing, output schemas are not documented, and error handling lacks recovery guidance. Tool 1 (devctx_whereami) has a clear, contextual description (~145 chars); Tool 2 (devctx_update_focus) is well-described (~210 chars) with good parameter coverage; Tool 3 (devctx_log) has a long description (~500+ chars) that mixes documentation with mandatory usage directives, indicating schema clarity issues. None of the tools document their return types or provide structured output schemas for the LLM to plan downstream steps. Schema validation is present (Zod types visible in code), but not exposed in tool definitions.
Log a development activity. Creates a timestamped entry in the activity log. **You MUST call this whenever you perform any of these actions:** - "build": After running any build command (npm run build, tsc, cargo build, make, etc.) - "run": After starting a dev server, running the app, or executing start scripts (npm run dev, ./start...
Update what you're currently working on. This sets the "current focus" shown in whereami and optionally updates the project description. Also syncs to CLAUDE.md (and AGENTS.md when present) so future sessions pick up the context.
Get a comprehensive overview of the current project state. Shows: current branch, git status, recent commits, active todos, branch notes, recent activity log, and current focus. Use this when starting a session, returning to a project after time away, or needing context on what's happening.
Output schemas not documented. Tools return complex nested objects (project state, git info, activity logs) but LLMs have no schema to understand expected fields, types, or pagination. Forces LLM to infer structure from undocumented responses.
Parameter descriptions incomplete or missing. Tool 1 input has one boolean param with description, but return structure (branch, git status, todos, activity, focus) is never documented. Tool 3 description embeds usage mandates ('You MUST call this...') instead of separating operational guidance from schema docs.
Error handling lacks recovery guidance. No error responses document what to do next (retry? ask user? alternate tool?). Errors from git operations, file I/O, or Linear sync are likely unguided.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 54 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 0 | - | v1 |
Tool 3 (devctx_log) name is vague. 'devctx_log' could mean retrieve logs, stream logs, or record logs. Verb is missing, 'record_activity' or 'log_activity' would be clearer. Pattern:tool requires verb_noun naming.
Tool 3 description is verbose (~500+ chars) and mixes operational requirements ('You MUST call this...') with schema documentation. LLMs struggle parsing mixed intent. Description should state WHAT it does, WHEN to call it, and WHAT it returns, separated from usage mandates.