Michael Ramos Skills

Edits prose to remove common AI writing patterns while preserving its meaning and intended tone.

Guides debugging by tracing symptoms to the first incorrect decision and correcting the source of the fault.

Applies a layered technical-writing standard to documentation, RFCs, READMEs, pull requests and commit messages for clear, precise prose.

Creates a project-local verification skill that launches, drives and checks an app through its real user surface, with evidence and cleanup.

Designs an auditable workflow for ambiguous, large-scale or multi-part engineering changes, with hypotheses, verification gates and decision logs.

Keeps a reviewable TSV decision trail for long-running, unattended or multi-phase work, including evidence and outcomes.

Removes unnecessary complexity before additions or refactors, trimming dead code, redundant checks and empty references before building on a simpler base.

Designs commands and processing loops that tolerate crashes, restarts and retries by converging on the same final state.

Turns non-trivial edits, analyses and checks into rerunnable scripts or other tools so the work is repeatable and reviewable.

Redesigns existing systems around new requirements by tracing affected files, rethinking foundational assumptions, and updating dependent references.

Refactors stateful logic around domain-fitting data structures to reduce branching, duplicated rules and invalid states.

Keeps reversible work moving by making reasonable decisions, presenting the result and reserving confirmation for irreversible actions.

Reviews code for reader load by tracing indirection and hidden state, then collapses needless layers and narrows mutable scope.

Guides software tests to assert user-visible behaviour rather than private implementation details, so they catch broken contracts and survive refactoring.

Creates or updates a personal working-style skill by reviewing workspace conversations, asking about preferences and drafting conventions for agents to follow.

Reconstructs recent working context from chat history, repository state and shared records, then returns current threads, problems and next steps.

Tests the shared assumption behind repeated failures before proposing another fix that depends on it.

Finds repeated agent mistakes in pstack and prevents them with architecture, types, lint checks, tests and targeted documentation.

Checks performance measurements for bottlenecks, fair tuning, errors, repeatability and real completed work before results are reported.

Checks whether a performance or evaluation result measures the intended work by identifying its limiter and ruling out alternative explanations.