superche/chestack

Use CheStack for evidence-driven engineering in Codex or ChatGPT: investigate, plan, build, debug, refactor, review, verify, or manage an authorized delivery workflow.

Design caller-facing interfaces, data ownership, types, and state transitions before implementation; deliver an implementable design package and revise it when implementation disproves its assumptions.

Inspect or resolve GitHub PR review and CI blockers until the requested stopping condition, without implicitly authorizing merge or future scheduling.

Compare candidate designs, implementations, or artifacts under shared criteria; choose a coherent base, synthesize useful alternatives, and verify the resulting artifact.

Drive, inspect, and profile a real CLI or TUI. Use for interactive prompts, keyboard flows, interrupts, terminal layout, startup regressions, or CLI hangs. Ordinary shell commands alone do not need this skill.

Drive and verify a browser, desktop, or Electron interface. Use for real UI reproduction, interaction acceptance, screenshots, accessibility checks, visual comparisons, or UI profiling. Source-only UI explanation does not require live control.

Create or update a reusable skill for Codex or ChatGPT, including triggering metadata, conditional references, portable tools, and validation.

Clean up code introduced by the current change. Use when asked to remove redundant or AI-generated code, or for a scoped cleanup before review or commit. Preserve behavior; a read-only review does not authorize edits.

Review or revise prose, documentation, and explanations for clarity, factual support, and preserved meaning. Use for editorial review, document cleanup, or removing empty AI-style prose; use code cleanup for implementation changes.

Explore an unclear problem or approach through bounded investigation and experiments. Deliver a supported proposal, experiment, or demo with unresolved production work; do not treat exploration as a production implementation request.

Trace current behavior and architecture from source when asked how a subsystem, API, state transition, or code path works. Distinguish source-derived behavior from runtime observations.

Recover working context across authorized history and current artifacts: goals, decisions, attempted fixes, live status, and next steps. Use for catch-up requests; resuming one known checkpoint uses session-pickup.

Review task evidence to improve skills, prevent recurring mistakes, or capture requested working conventions.

Review a diff or design for correctness, regressions, test gaps, type boundaries, and effects beyond changed lines.

Set up or configure CheStack when the user asks to install it, inspect its available capabilities, or change its workflow preferences. Do not use for ordinary application setup or debugging.

Teach a subsystem, change, or concept by combining grounded how and why findings into an explanation matched to the reader. Use for onboarding and requests to really understand, not to implement.

Verify real behavior, create a project-local verification skill and feature map, or maintain an existing verification harness against source and live behavior.

Investigate historical rationale, design decisions, thresholds, regressions, and tradeoffs using source history and authorized records. Separate documented intent from inference and missing evidence.