Claude Skills by hani-q

Resolve material open questions in a QStack plan through one-at-a-time, plain-language decision dialogs, then write each answer and its consequences back into the plan immediately. Accepts a bootstrap Markdown plan or an authoritative QStack plan.html; once HTML exists, edits only HTML and never maintains a Markdown mirror. Use when invoked as /qstack-ask-plan-open-questions, when asked to resolve a plan's open questions, or automatically after /qstack-plan-to-html creates plan.html.

Rewrite the immediately previous assistant answer in plain, non-technical language using far fewer lines. Use when the user invokes `/qstack-be-concise`, optionally with a target such as `/qstack-be-concise 4`, or asks to make the last answer shorter and simpler.

Assess what a proposed or recent change could break beyond its direct diff, identify the critical safety fact, and prove it with real code or mark it unproven. Use for blast-radius analysis, "what could this break?", or a small change whose downstream effects are uncertain.

Build the smallest rerunnable tool that makes non-trivial work repeatable, reviewable, and safe to verify.

Turn a recurring correction or repeated instruction into the strongest practical structural guardrail. Use when the same mistake returns, prose is being repeated, or a rule could become a type, schema, lint, check, or script.

Rewrite the immediately previous assistant answer for a named reader, such as a product manager, CEO, or CMO, optionally framed around a named product. Use when the user invokes `/qstack-explain-for [role] [product]`, says the last answer is too technical, too low level, or too narrow, says they do not understand it, or asks for it at a higher level, in simpler words, or from a role's or product's point of view.

Reproduce and diagnose a failure until its root cause is supported by direct evidence. Use for debugging, recurring symptoms, restart-only failures, or a proposed guard that may only hide the real defect.

Settle data shape, access, ownership, concurrency, and useful shared setup before writing feature logic.

Explain how a subsystem works by tracing its real runtime flow, ownership, boundaries, and files. Use for code walkthroughs, onboarding explanations, placement or layering questions, and explicitly requested architecture critique.

Catch the code that got built because it seemed useful rather than because the goal required it. Restate the goal, test each assumption against evidence, then remove, simplify, optimize, or automate in that order and never further than the goal needs. Does not cut validation, error handling that prevents data loss, security, or a repository invariant. Use as /qstack-libero after a task or feature is built, or when asked what can be deleted or simplified.

Execute an approved implementation plan exactly, maintaining QStack execution evidence and asking how much independent adversarial review to run.

Execute an approved implementation plan with documented adaptations, QStack execution evidence, and a user-selected adversarial review depth.

Design or review commands, lifecycle steps, and processing loops so retries, partial failure, restart, and stale state converge to the same correct result. Use when an operation may run twice or resume after a crash.

Replace scattered domain assumptions with the smallest structure that removes invalid states and branching. Use when booleans have to be kept in sync, the same rule is validated at several call sites, a branch grows with every new case, or an invalid state can be represented at all.

Recommend the one installed skill to run next, from what the conversation has done so far and the state of the plan folder on disk. Use when invoked as /qstack-next, when asked "what now?" or which skill to use next, or as the last step of /qstack.

Review whether an approved QStack HTML plan was faithfully executed by comparing plan.html with execution.md, executor.md, or legacy implementation notes and the actual committed and uncommitted code changes. Build a requirement-to-evidence matrix, identify missing or unapproved deviations, check whether the execution record matches reality, and assign a guarded 0-5 adherence score. Use when invoked as /qstack-plan-adherence-review, when asked whether a plan was followed, or before closing or...

Close an executed plan by recording its outcome against the plan and the execution board.

Read the project's existing plan folder as content and brief the planner on what was already decided, deferred, and learned. Runs standalone before anything is written, or from /qstack-plan-to-html once that skill has settled its source and before the HTML document is. Ranks earlier plans by overlap with the subject being planned, then reports decisions already settled, questions never answered, where plans diverged from reality, live board cards owning the same files, what this plan supersed...

Render a plan as a reviewable QStack HLD/LLD HTML document, resolve its material open questions, and break it into the cards of its execution board. Takes a Markdown draft or the working agreement reached in the conversation. Also adds a board to a plan that already has HTML and no board.

Verify completed work against the real artifact and its full input-to-output path instead of relying on compilation, timestamps, cached output, or agent summaries. Use before claiming a task or feature works.

Produce an evidence-only report on how a project is actually being worked: checkout and branch topology, momentum over time, rework, instruction-file churn, and plan-record completeness. Every finding cites a count the reader can reproduce, and any category without enough evidence is refused rather than padded. Reads across every worktree Git tracks for the repository, not just the current checkout. Use when invoked as /qstack-reflect, optionally with one or more plan directories such as /qst...

Review a pull request, a branch against its base, or the working tree against a generic correctness baseline plus the repository's own CODE_REVIEW_RULES.md, then report findings and a score computed from their severities. Next to /review and /code-review, this one adds repository rules files, a score no reviewer is asked to choose, and a report-only contract: nothing is edited, posted, or landed.

Redesign concurrent work that shares a mutable file, branch, key, or state object by separating ownership before adding serialization. Use when actors may race, a lock is proposed, or coordination relies on taking turns.

Serve the current repository's QStack HTML plan collection on a chosen local address and port.

Build a clickable static HTML prototype of one or two screens and keep it in the repo under qstack/compound_engineering/prototypes/<slug>/, or attach it to a QStack plan, where it is embedded in plan.html through an iframe with a link to open it in its own tab. Works from a brief in the conversation, a design doc, or an existing screen in the code. No framework, no build, no network. Use when invoked as /qstack-ui-prototype, when asked for a prototype, mockup, or wireframe that should be kept...

Rewrite the previous answer to remove AI writing patterns and add a natural human voice without losing meaning or useful detail.

Refresh a repository's shared QStack plan template from the installed qstack-plan-to-html template without rewriting any plan or execution record. Use when plan board styles or scripts are stale, when qstack/compound_engineering/plans/.template/v1 differs from the installed skill, or automatically from /qstack-plan-to-html after it finds an older shared template. Requires qstack-plan-to-html, which owns the source.

List every skill installed from QStack and from the optional collections QStack installs (Matt Pocock's skills, human-review) as a table with a one-line description each, then recommend the next skill to run from the conversation so far and the plan folder on disk. Use when invoked as /qstack or when asked which QStack commands exist.