Comment.io overlay for Matt Pocock's skills

A thin explicit wrapper for mattpocock/skills:

• task context comes from the conversation and repository; no task tracker is required;
• specs, plans, ADRs, research, handoffs, and other optional artifacts live in Comment.io;
• implementation closes with one risk-scaled review lane; Matt's review replaces the general lane only when a line-addressable spec needs traceability or a non-tool-enforced standards audit is an acceptance concern. Every candidate gets simple, startup-pragmatic judgment before it becomes work.

Matt's skills still own the workflow. This overlay is runtime-generic: the same skills work in Claude Code, Codex, Cursor, Gemini CLI, and other Agent Skills clients.

Install

Install Matt's skills first, then this overlay:

npx skills add mattpocock/skills
npx skills add comment-hq/skills

The replacement review uses fresh native subagents with read/search access to an immutable target. Use a host with native delegation; Grok Build plus jq is an optional explicitly selected second opinion or an alternative for hosts without subagents. See the bundled code-review skill for the exact workflow and limits.

The order is intentional: this overlay replaces Matt's code-review in the global skill namespace. Matt's original remains available as mattpocock/skills@code-review.

Engineering delivery uses the supplied task context and preserves explicit approval, review, and delivery safeguards. A worklog is optional evidence. If the comment skill is installed, the overlay invokes it so it can reuse or mint the session identity. It invokes listen too when that skill and the current runtime support idle mention delivery.

Invoke comment-dev alone to start with ask-matt, or supply a target Matt skill to run it directly:

# Codex
$comment-dev
$comment-dev $grill-with-docs <idea>
$comment-dev $implement <task context>

# Claude Code with copied skills
/comment-dev
/comment-dev /grill-with-docs <idea>
/comment-dev /implement <task context>

The bundle contains comment-dev, the replacement code-review, and its model-invoked review-judgment helper.

License

MIT — see LICENSE.