
Turn a CHANGELOG entry (or set of shipped commits) into a launch post draft for Twitter, a blog, or an email. Bridges engineering output to marketing output without requiring a marketing agent.
Write a lightweight design doc before implementation for Medium/Large tasks. Mandates Non-Goals and Alternatives Considered sections to prevent scope creep and hindsight bias. Inspired by Google design doc culture.
3-line ADR lite. Capture a decision you just made in under 60 seconds — problem, choice, rejected alternative. Complements the heavier `adr` skill; use this when a full ADR is overkill but the decision will be forgotten.
Pre-merge and pre-deploy checklist for solo-founder projects. Absorbs the former devops agent's gate role as a lightweight skill. Use before squash-merging to main and before promoting a build to production.
Execute an implementation plan step by step with verification at each step. Use when implementing code from an approved plan.
Structured code review with checklist covering logic, security, maintainability, and tests. Use when reviewing code changes before merge.
Pipeline retrospective and pattern extraction for continuous learning. Use after every pipeline completion to capture lessons.
Create and manage git worktrees for isolated development. Use when pipeline needs an isolated workspace.
Elicit and structure requirements through systematic questioning. Use when a feature needs to be scoped before design.
/review Layer 3: Main Agent responds to every Layer 2 finding with a disposition. Hard cap 2 needs-user-decision per /review run.
Explore and understand codebase structure, patterns, and conventions. Use before making design decisions or diagnosing bugs.
Enforce pipeline progression rules — hard gates require user approval, soft gates auto-proceed on success. Use whenever a pipeline stage has a gate annotation.
Write blameless postmortems for significant bugs. Use after a bug fix to capture lessons learned.
Weekly shipping ritual for solo founders. Scan the last 7 days of commits, draft a changelog entry, and pick next week's single top priority. Inspired by Duolingo's "Ship It" culture. Invoked via /ship-weekly.
Force every feature spec to answer "how will we know this worked?". A single-field skill that gates feature requirements. Inspired by Duolingo's metric-driven culture — the solo-dev version of A/B testing discipline.
Sketch low-fidelity UI wireframes before frontend work. Absorbs the former ui-designer agent's role as a skill. Uses Pencil MCP if available; otherwise produces an ASCII/markdown wireframe.
Command pre-step: load REQUIREMENTS/DESIGN/TODO + classify MEMORY into BRIEF.md. The only skill allowed to read MEMORY.md.
Run and analyze test results against acceptance criteria. Use when verifying implementation completeness.
Scan code changes for security vulnerabilities. Use as part of code review or independently.
Systematic root cause analysis for bug diagnosis. Use when investigating the underlying cause of a bug.
Company of One v2 orchestrator. Read this skill ONLY when a user message implies work (bug, feature, refactor, plan, ship) AND the task is Medium or Large. Do NOT read for questions, explanations, or Small tasks.
Git branch, commit, and merge operations following project conventions. Use for all git workflow actions.
Write Architecture Decision Records for significant technical decisions. Use when a decision will affect future development.
Write detailed implementation plans at file-level specificity. Use when breaking a design into executable steps.
Post-code check: run tests + confirm TODO acceptance items map to passing tests; applies a security lens but is not a separate scan.
/dev post-step: 9-question retro including Q9 prediction accuracy (ADR-002). Triggers reconstruction drill only for diverged core items.
/debug step: test one hypothesis at a time. Hard rule: no shotgun debugging — no concurrent changes across hypotheses.
Author an Architecture Decision Record under docs/projects/<project>/adr/. Only when a decision has cross-feature, immutable, long-lived consequences.
Reverse-question the user until the feature intent is unambiguous. Drives brief Open Questions to empty before downstream skills run.
/debug step: enumerate 3–5 plausible root causes with explicit uncertainty labels. No padding.
Split a feature into 60% boilerplate vs 40% Human-Owned Core; force the user to predict core behavior before Execute. Gates /dev per ADR-002.
/review Layer 2: adversarial review by reviewer agent reading ONLY REVIEW_INPUT.md. Three output buckets; security is a lens, not a separate skill.
External-knowledge gathering step — Web Search + Context7 docs lookup. Validates that the spec/plan reflects current library reality.
/review Layer 1 auditor + /think self-review. Three contexts: think-self-review, review-spec, review-implementation. Also judges reconstruction drills.
Author REQUIREMENTS.md + DESIGN.md + TODO.md for a feature. The three files are one contract; they ship together.
Test-driven development cycle: RED → GREEN → REFACTOR. Every code change goes through it, including /debug regression tests.
Decide WHAT to test before writing tests. Output: golden / edge / error / won't-test. Feeds tdd.
Command post-step: write CHANGELOG + TODO once per command run. One call, not per-skill doc writes.
Update BRIEF.md sections during a command run. Any skill that produces a brief-persisted artifact calls this to write it back.
Pricing heuristics for solo-founder SaaS products. Use when designing or revising a pricing page, choosing a tier structure, or deciding between usage-based vs seat-based billing. Outputs a concrete pricing recommendation, not a lecture.