← ClaudeAtlas

explorelisted

Read-only deep exploration and skeptical gap analysis before any code is written. Use at the start of every non-trivial task, when onboarding to an unfamiliar codebase or module, or when the user asks "how does X work here", "analyze before changing", or requests an As-Is/To-Be gap analysis.
omeeragtoprak/agentic-engineering-protocol · ★ 0 · AI & Automation · score 72
Install: claude install-skill omeeragtoprak/agentic-engineering-protocol
# AEP Phase 1 — Explore (read-only) **Hard rule: no production code, no file edits, no destructive commands in this phase.** Your only outputs are knowledge and a gap analysis. ## 1. Ingest Read before reasoning — in this order of leverage: 1. The files named in the task, plus everything they import/depend on (one level out). 2. Project instruction files (AGENTS.md / CLAUDE.md §P), lockfiles, build configs, CI definitions, lint/format rules. 3. Existing tests around the affected area — they encode the real contract. 4. Recent git history of the affected files (`git log --oneline -15 -- <path>`) — it encodes intent and churn. Match the repository's established conventions exactly; note them, don't fight them. For wide investigations (many files, unfamiliar subsystems), delegate to a subagent or a scoped search so the main context stays clean — then spot-check the findings against the actual files before relying on them. ## 2. System map (for changes crossing module boundaries) Before editing anything that other code depends on, build a small **impact map — mechanically, not from memory** (guessed maps encode the same blind spots that cause the bug): ``` ## System Map Callers: <who invokes the touched symbols — from grep/LSP references, not recall> Callees: <what the touched code invokes across module boundaries> Data flow: <where data enters/leaves the touched code: params, globals, DB, files, events> Co-change: <files that historically change together — git lo