← ClaudeAtlas

staged-coding-1-reconlisted

staged-coding stage 1 - recon (read before you write). Traces the real call path, checks live database state in every environment, and forces the caller list, live data and the ticket's instruction into the evidence before any hypothesis is allowed. Signs its ledger line on finish. Triggers: /staged-coding-1-recon, or invoked in order by /staged-coding.
Shadysmetools/claude-skills · ★ 0 · AI & Automation · score 72
Install: claude install-skill Shadysmetools/claude-skills
# Stage 1 - recon (read before you write) Part of the `/staged-coding` chain. Runs after Stage 0 (chase) and before Stage 2 (restraint). Callable alone (e.g. to re-run recon on someone else's PR); the conductor carries the cross-cutting rules. ## Inputs - The ticket (or, for an epic, every slice) and its lane, from Stage 0 - The repo's architecture map, if it has one (e.g. `docs/architecture/CODEBASE-MAP.md`) - Read-only access to live data: `config.database.read_only_query_cmd` (or an equivalent MCP) for each environment, plus any app-level read API - Your notes on each external vendor the change touches (how it is wired for you, which keys and clients it uses, and known past defects with that vendor). Read them first if they exist: a known defect class is cheaper to look up than to re-derive. ## What to do **Trace the real call path** of whatever you are about to touch. State plainly what already exists before proposing to build anything new. Reuse beats build: a second parallel implementation of something that already exists is a classic "dual-system" bug. **Live-state READ.** Source and docs describe intent; they do not describe what is stored. Before assuming the defect lives in the code, look at the real data the ticket is about (query the relevant tables read-only) and the real server state (env config, a running process, a queue depth) if either is plausibly involved. Sometimes the problem isn't in the code: the numbers in the database don't match what the