Thomas Roche
Community@thoroc · Edinburgh
Thomas Roche maintains 19 skills for structured engineering documentation: ADRs, implementation plans, risk registers, tech debt ledgers, and decision-review workflows.
Agent Skills by Thomas Roche
Showing 19 vetted skills indexed across 2 GitHub repositories.
thoroccontext-file
Create structured .context/ Markdown files with standard YAML frontmatter for plans, findings, and analyses.
thorocdesign-debate
Stress-test unwritten decisions by spawning opposing subagent roles and synthesizing a validated verdict.
thorocqmd
Search markdown knowledge bases using BM25, vector, and hyde query types.
thorocdocs-check
Validate docmd-built documentation sites through build, orphan, ADR index, and LLM output checks.
thorocrisk-register
Maintains an append-only Markdown risk register tracking deferred items, shortcuts, and open decisions.
thoroctech-debt
Maintains a schema-validated Markdown table tracking low-stakes code cleanup items in docs/TECH_DEBT.md.
thorocguided-interview
Conduct adaptive one-question-at-a-time interviews with concrete options and confirmed recaps.
thorocadr-capture
Extract binding decisions from context files into numbered, immutable Architecture Decision Records.
thoroccontext-index
Regenerate and validate a YAML index of .context Markdown files from their frontmatter.
thorocsession-reflection
Conducts a two-question session-end reflection to surface confidence gaps and blind spots.
thoroctech-evaluation
Evaluates libraries and file formats against cited evidence to produce one schema-validated recommendation.
thorocplan-create
Create structured .context/plans/ Markdown files with validated YAML frontmatter and phased task decomposition.
thorocrules-management
Manage project-level agent behavioural rules in a single schema-validated rules file.
thorocplan-review
Reviews plan files using three parallel subagent reviewers with technical, strategic, and risk lenses.
thorocfollow-up
Reports open follow-ups and recommends the next backlog item without mutating files.
thorocmermaid-diagrams
Generate validated Mermaid diagrams from code and natural language descriptions.
thorocacceptance-criteria
Craft acceptance criteria using Given/When/Then and checklist formats.
thorocconventional-commits
Enforce Conventional Commits v1.0.0 for automated changelogs and versioning.
thorocmoscow-prioritization
Categorize project requirements into Must, Should, Could, and Won't Have.
Frequently Asked Questions About Thomas Roche
FAQPage SchemaWhat tasks can I accomplish with thoroc's skills?▼
You can create and review implementation plans in .context/plans/, capture Architecture Decision Records, maintain risk registers and tech-debt ledgers, generate validated Mermaid diagrams, write Given/When/Then acceptance criteria, enforce Conventional Commits, apply MoSCoW prioritization, and search markdown knowledge bases via QMD.
Who are these skills designed for?▼
Software engineers, technical leads, and architects who maintain structured project documentation. They suit teams practicing ADR-driven development, conventional commit versioning, and disciplined tracking of deferred work, technical debt, and open architecture decisions across a repository's docs/ and .context/ directories.
How do the plan creation and review skills work together?▼
plan-create scaffolds .context/plans/*.md files with YAML frontmatter and phase/task decomposition, inferring local style from existing plans. plan-review then runs three independent subagent reviewers — Technical, Strategic, and Risk — with user-selected models, consolidating their structured feedback into one assessment.
What are the licensing and dependency requirements?▼
The qmd skill is MIT-licensed and requires the qmd CLI or MCP server, installed via `npm install -g @tobilu/qmd`. Other skills operate on repository files (.context/, docs/ADR/, RISK_REGISTER.md) and rely on the host agent environment; plan-review supports OpenCode Zen, OpenCode Go, Anthropic CLI, or BYOK model backends.
How do the risk-register, tech-debt, and follow-up skills differ?▼
risk-register logs deferred items, shortcuts, and open decisions with real cost if unaddressed; tech-debt tracks code-level cleanup like lint violations; follow-up is a read-only status check listing active entries and recommending the single best next item, pulling from both registers when the follow-up list is empty.