← ClaudeAtlas

coderail-framelisted

Investigate repository evidence, detect candidate domain boundaries, and choose a quick path or one high-impact novice-safe question before drafting a CodeRail contract. Use internally for new, ambiguous, risky, persistent, or cross-module requests whose task boundary is not yet established.
HaipingShi/coderail · ★ 5 · AI & Automation · score 67
Install: claude install-skill HaipingShi/coderail
# CodeRail Frame Frame the next bounded slice before contract drafting. Reduce decision uncertainty without making the user answer questions the repository can answer. ## CodeRail boundary - Read `AGENTS.md`, `docs/NORTH_STAR.md`, `docs/TASKS.md`, relevant decisions, and `git status` before framing the request. - Treat this skill as a read-only reasoning primitive. Do not create a task, edit files, implement, or persist the frame. - Do not write `.coderail/tasks.json` or edit task state behind the CLI. - Never stage or commit task changes. - Any later implementation finishes through `python .coderail/coderail.py done`. ## Frame 1. Extract the requested outcome and the repository facts already supplied. 2. Generate at most three domain lenses that could materially change `G`, `T`, `S`, `V`, `X`, or `P`. Label each as `[CANDIDATE]`. A candidate lens is never a fact about the product or user. 3. Investigate the repository before asking anything. Inspect instructions, code, tests, configuration, decisions, and history relevant to each lens. 4. Type every material claim: - `FACT`: cite repository, command, request, primary-source, or experiment evidence. - `ASSUMPTION`: name the owner, falsifier, and risk. Point high-risk assumptions to a proposed `X` stop condition. - `DECISION`: name the decision owner. - `UNKNOWN`: mark it exactly one of blocking or deferred. 5. Remove lenses that repository evidence resolves. When repository evidence