Skip to content

Comparable projects

An explanation. Other people are solving adjacent problems. This page names them, states what each targets, and says plainly how procoder's approach differs. It is descriptive, not competitive.

It is deliberately separate from Influences, which records what procoder took from other tools. Some projects appear on both pages, and where they do, this page is not the place the debt is acknowledged — that page is.

What this page can and cannot argue

There is a tempting argument that several unrelated teams building the same defences is itself evidence the problem is real. It is a good argument, and it does not apply to most of what is listed here.

Procoder read unlazy, addyosmani/agent-skills and mattpocock/skills in a research sweep on 2026-08-26, and shipped mechanisms from all three within days. Those are borrowings. Calling them convergence would be a claim this repository's own issue bodies contradict, on a page whose whole purpose is to be checkable.

The narrower claim does hold, and it is worth stating. Procoder's commit gate landed on 2026-08-18 and its evidence-gated closes on 2026-08-19 — both before that sweep, in a project whose first commit was 2026-08-16. unlazy independently ships acceptance ledgers with runnable gates and required evidence. Two projects reaching "a claim of completion is not evidence of completion, so make the evidence runnable" without contact is a real signal about the failure mode, and it is the one convergence claim on this page that survives its own check.

The projects

unlazy

Targets: model laziness, underthinking, and premature completion claims, at the level of a single task's acceptance ledger.

Overlaps: runnable verification gates with recorded evidence; a Stop hook that blocks a turn ending on incomplete work; parallel-dispatch honesty; ownership leases for concurrent agents.

Diverges: unlazy is task-scoped and ledger-centric — the unit is one piece of work and its gates. Procoder is repository-scoped: the same honesty applied to a commit, a sprint, a release, and ten domain checks that run whether or not any task is open. unlazy also executes approved checks on re-verification; procoder's binary executes nothing it was not explicitly asked to, and modifies no file outside its own cache.

Also: its research/validation-protocol.md retracts an earlier internal benchmark rather than quietly dropping it, which is the discipline Research is held to.

addyosmani/agent-skills

Targets: the full engineering lifecycle as 24 skills, plus specialist personas and reference checklists, installable across 70+ agents.

Overlaps: per-skill rationalization tables against agents talking themselves out of steps; progressive disclosure to keep entry-point files small; multi-angle review as separate personas.

Diverges: it advises and procoder computes. A skill telling an agent to run the tests and a binary refusing to close the task when they did not run are different kinds of object. It also ships an honest docs/comparison.md naming its own alternatives, which is the model this page follows.

mattpocock/skills

Targets: disciplined workflows — TDD, code review, debugging, domain modelling — as model- and user-invoked skills.

Overlaps: its wizard skill scripts a walkthrough for steps only a human can perform; its handoff compacts work for the next agent, which procoder keeps as .procoder/state/handoff.md.

Diverges: same advice-versus-computation line as above.

anthropics/skills and the Agent Skills spec

Targets: the shared format itself, rather than a competing opinion.

Relationship: procoder conforms to it. skills/procoder/SKILL.md is a spec-compliant skill and plugin.json a spec-compliant plugin. This is the foundation both this page and Influences are written on top of, and naming it is more useful than leaving it implicit.