SKILL CATALOG
Workflows, not prompt tricks.
Each skill owns a bounded engineering activity, states when to stop, and keeps its evidence and authority limits visible. Use one directly or let your agent select a matching path.
01
Project foundation
Establish the repository-owned engineering contract without adding a Baseline runtime or overwriting stronger local rules.
Start with $setup-baseline
02
Change workflow
Move from a governing request to a verified change with depth selected by risk, not ceremony or line count.
Start with $measurer
$measurer Classify risk and route only the engineering depth the work needs.
$refine Resolve materially incompatible interpretations without reopening clear decisions.
$tdd Implement approved behavior through the smallest fail-first executable check.
$bugfix Repair an existing defect from reproduction to causal fix and regression proof.
$verify Review the governing input, complete diff, evidence, risks, and limitations.
$git-commit explicit Create one safe atomic local commit from a verified task-owned slice.
$ci-workflow Build or repair CI that produces traceable evidence with least privilege.
$docs Preserve only the knowledge that must outlive the current task.
03
Design and architecture
Align language and boundaries, compare material options, or inspect architecture without turning routine work into ceremony.
Start with $shape-domain
$shape-domain Align domain language, invariants, scenarios, and ownership boundaries.
$design-deep-modules Design high-leverage boundaries with small interfaces and hidden complexity.
$improve-architecture explicit Audit a codebase for evidence-backed architecture improvements.
$decision-framework Choose among viable paths with evidence, uncertainty, and revisit triggers.
04
Explicit deep work
Open a deliberate exploration, forecast failure, investigate current evidence, or hand work to another session.
Start with $brainstorming
$brainstorming explicit Explore materially different directions before selecting one.
$premortem explicit Forecast plausible failure chains and make mitigations observable.
$session-bridge explicit Create a truthful compact handoff another session can resume.
$technical-research explicit Answer a technical question with primary sources and explicit limits.
05
Safety
Inspect security, privacy, data-loss, trust, and authority boundaries without turning declarative guidance into an enforcement claim.
Start with $security-review
KEEP THE BOUNDARY CLEAR
Skills guide work. Your environment enforces policy.
Declarative instructions do not enforce chronology, scope, sandboxing, permissions, branch protection, or production authority. Those controls remain authoritative where configured.
Read the methodology