factory-dashboard
claudeA pane with the factory's numbers: what's waiting on you, the loop's spend and rework, evals and tokens
Stelliad's skills, pinned to a reviewed commit. Move the pin by editing sha after reading the diff
Plugin stelliad from marketplace sanvio-factory in SanvioLabs/sanvio-factory. Commands read from the repo's manifest — yesterday.
ai-contract-auditAudit whether your contracts match how your team builds with AI, disclosure, training posture, IP assignment, confidentiality, warranties, regulatory.
anonymizeProduce a shareable copy of a file, folder or repository with client names, people, account IDs, domains and secrets consistently replaced, then check the copy for anything the pass missed.
compose-from-interviewInterview the author for the stories, numbers, and opinions a draft needs, then write from their answers rather than inventing the substance.
compose-scoreScore a draft on seven dimensions against a stated bar, including whether its diagrams and screenshots earn their place, fix what is editorial, and route what is missing back to the author instead of generating it.
coverage-gapsFind the code paths no test reaches and the external calls no test mocks, ranked by the risk of the code that is uncovered.
create-specScaffold a new feature spec from an idea, writing SPEC.md into its own numbered folder (default specs/{NNN}-{slug}/) from the shipped template, with what was actually said filled in and what was not left open as questions. Stops there, as a draft; no plan, no tasks, no code. For a codebase that alre
create-ticketTurn a rough idea into a ticket that passes the Ready gate, or split an oversized one. Inspects the repo for real context, applies the Ticket Authoring Standard, records assumptions instead of inventing facts, and calls review-ticket before declaring anything Ready. Use when saying "make a ticket fo
design-printed-partDesign a parametric CadQuery part or enclosure meant to be FDM printed, and hand over print-ready output. The model proves its own fit by measured interference before any filament is spent, and the print differs from it only where printing demands.
find-dead-codeFind orphaned files, unused exports, dead dependencies and commented-out code, and report them with a confidence level rather than deleting anything.
generate-readmeGenerate a README from what a codebase actually contains, rather than from a template with placeholders left in it.
generate-svgHand-craft production-ready SVG images and animations from a description: clean paths, a real viewBox, CSS-only motion, and no embedded rasters.
implement-specWork a spec's task list forward one task at a time, test first and verified, updating the checkbox, the evidence and the requirement traceability as it goes, and stopping when the spec turns out to be wrong.
plan-specRead a ready SPEC.md, inspect the repository it governs, and write PLAN.md and TASKS.md beside it, every task carrying its requirement IDs, its dependencies and a verification command that actually exists in the repo. Never modifies application code. Halts and reports the gap if the spec is not read
recover-specWork backwards from an existing codebase to the specification nobody wrote. Fixes the system boundary, inventories every surface, traces the core data path end to end, recovers the data model and its unenforced invariants, and reads the tests as the only stated requirements the system has. Grades ev
repo-auditAudit a repository against a security and quality baseline, and report which gaps block the delivery phase it is in.
review-intakeProcess code review feedback as a technical exchange: read the whole set, clarify before touching a file, verify each item against the codebase, disposition every one, and push back with evidence where the reviewer is wrong.
review-principlesReview a codebase against SOLID, DRY, and KISS, and report findings tied to a consequence rather than to taste.
review-skillReview a skill from outside your harness before it gets anywhere near your agent. Quarantine it, read every file, flag risks, sort each behaviour into covered, gap or conflict against what you already have, and draft only the gaps in your own conventions. Use when asked to install, add, import, try
review-specIndependent final review of a spec's implementation against SPEC.md, PLAN.md, TASKS.md, the diff and the tests. Walks every requirement to the code and test that meet it, delegates the lenses to reviewers that already exist (your harness's code review, review-principles, find-dead-code, coverage-gap
review-ticketGrade a ticket against the Ticket Authoring Standard and return READY, NEEDS_REFINEMENT, or BLOCKED with a score out of 12, naming exactly what stops safe execution. Also reports when a ticket holds more than one deliverable outcome and how it splits. Reports only, never edits the ticket. Use when s
run-gatesCheck a repository's gates before a merge, a deploy or a phase change by running the conditions in its .gates.yaml (files that must exist, patterns files must contain, commands that must exit clean, and other gates) and reporting pass or blocked with the fix for each failure.
secret-scanScan a repository for exposed credentials, map the blast radius of each one, and produce rotation guidance ordered by what an attacker reaches first.
stress-test-planInterrogate a plan or decision one round of questions at a time, working a design tree from the questions that can be answered now, until the weakest assumption is named.
tech-doc-reviewScore a technical document on claim clarity, structural completeness, precision, defensibility, novelty, and slop detection. Loops revision until it hits a 9/10 bar. Distinguishes between editorial fixes (the machine handles) and knowledge gaps (routes back to the author).
test-firstEnforce the red-green-refactor cycle: require a failing test before the code that satisfies it, and halt when coverage falls.
third-party-complianceThird-party compliance checkpoint. Assess whether a third-party app or service is safe to use under a regulatory framework (HIPAA, SOC 2, COPPA, GDPR, or FERPA), or assess your whole stack at once by discovering every vendor a codebase actually talks to. Produces a client-shareable compliance summar
triage-alertIngest a monitoring alert, trace it to the commit that introduced it, and route it to one of four outcomes: fix, file, escalate, or suppress.
verify-doneGate every completion claim behind fresh evidence: name the command that proves it, run it, read the output, then state the claim with the result attached.
A pane with the factory's numbers: what's waiting on you, the loop's spend and rework, evals and tokens
A pane showing the loop in focus: what's waiting, what each issue is on, and its pull request
A line above the prompt with the repo's running actions, open pull requests and open issues
A line above the prompt showing which part of the harness just acted
Matt Pocock's agent skills for real engineering (grilling, spec/ticket flows, TDD, code review, domain modelling and more).
39 skills
Up-to-date documentation lookup. Pull version-specific documentation and code examples directly from source repositories into your coding agent's context.
1 MCP server
Automates browser interactions for web testing, form filling, screenshots, and data extraction
12 skills
Use Alibaba Open Code Review from local Codex.
2 skills
Backend API design, GraphQL architecture, workflow orchestration with Temporal, and test-driven backend development
9 skills · 1 command · 8 agents
Search and retrieve documents from local markdown files.
2 skills
This page updates itself from your repo. Add the badge to your README, or get it Featured to pin it above the organic results with a followed link to your site.