Skip to content

cometweb-io/cometweb-agent-skills

v2.0.1MIT

CometWeb Labs specialist skills for research, product operations, quality workflows, adversarial review, QA, release readiness, and evidence-based decisions.

CometWeb Agent Skills

Reusable skills that help AI coding assistants research a question, choose the next product task, audit a workflow, and review a release.

A request passes through a specialist skill to a structured result; tools and approvals stay with the host.

Validate MIT Skills

32 skills · Cursor, Claude Code, Codex and compatible hosts

This repository contains 32 reusable skill packages for compatible agent hosts.

Start here

git clone https://github.com/CometWeb-io/agent-skills.git
cd agent-skills
./scripts/install-codex.sh

Use ./scripts/install-claude.sh for Claude Code or ./scripts/install-cursor.sh for Cursor. For all supported hosts, use ./scripts/install-all.sh. See installation and host setup.

Then ask your assistant:

Use product-operator to compare this repo with the roadmap.
Give me the three most useful next actions and how to verify each one.

The assistant loads the skill's instructions and uses the tools available in your host. Browser access, connectors and external actions depend on that host and your authorization.

Pick a starting point

You need to…Start withYou get
Check a claimevidence-researcherSources, contradictions and an Evidence Pack
Decide what to do nextproduct-operatorA bounded now / next / stop list
Inspect an appweb-app-auditorFindings tied to observed behavior
Review a releaserelease-readinessA verdict for a specific candidate
Combine specialistsskill-orchestratorOrdered steps and structured handoffs

For isolated specialist runs, use skill-orchestrator-multiagent.

Browse all 32 skills

Skills

Foundation and orchestration

SkillUse it for
ai-councilEvidence-governed decisions, risk gates, forecasts, and GO / TEST / DEFER verdicts.
cometweb-contextFresh, provenance-aware context snapshots before work that depends on current project state.
evidence-researcherClaim decomposition, source verification, falsifiers, contradictions, and Evidence Packs.
portfolio-operatorCross-project focus, capacity conflicts, and pause / delegate decisions.
skill-orchestratorMulti-skill workflows with ordered steps and CW-AIP handoffs.
skill-orchestrator-multiagentIsolated subagent execution for multi-skill workflows.
benchmark-curatorBenchmark corpora, holdouts, contamination controls, and revision hashes.
feedback-integratorRecurring failure patterns, improvement proposals, and regression tests.
quality-loop-operatorBriefing, review, repair, acceptance, measurement, and quality lifecycle control.
rubric-designerObservable evaluation criteria, evidence floors, blocker rules, and rubric locks.
skill-auditorSkill routing, portability, package hygiene, and supply-chain audits.
skill-evaluatorFair skill experiments, behavioral lift, resource cost, and host comparisons.

Product, research, and partnerships

SkillUse it for
ai-humanizeNatural English and Polish rewrites that preserve meaning and voice.
competitive-intelligenceCompetitor watchlists, change detection, and recurring delta digests.
design-partner-finderFinding, qualifying, and managing design partners and early adopters.
product-operatorWeekly product control loops, roadmap drift, and now / next / later / stop actions.
product-teardownEvidence-backed product, UX, architecture, and implementation pattern analysis.
repo-to-roadmapWhole-project baselines, gap inventories, dependencies, and target-state roadmaps.
brief-architectExplicit artifact contracts, evidence policies, risks, and acceptance criteria.
content-writerEvidence-aware reader-facing articles, guides, reports, and documentation.
content-reviewerConstructive editorial QA with evidence-backed, actionable findings.
content-roasterAdversarial content review, proof debt, objections, and repair verification.
science-roasterReviewer #2-style critique of methods, inference, validity, and reproducibility.
repo-roasterAdversarial repository review with invariants, reachability, and repair contracts.
repair-operatorMinimal dependency-aware repairs and fresh verification of closed findings.
artifact-acceptanceFinal evidence-backed acceptance gates for knowledge artifacts.

Publication, operations, QA, and release

SkillUse it for
customer-opsSupport triage, incidents, account risk, commitments, and engineering handoffs.
ebook-publisherResearch-backed ebooks, white papers, workbooks, and publication QA.
longform-publisherCanonical long-form manuscripts and release-ready derived documents.
release-readinessCandidate-bound production gates and GO / GO_WITH_CONTROLS / NO_GO / DEFER verdicts.
seo-geo-aeo-maxxingMulti-pillar SEO / GEO / AEO visibility audits.
web-app-auditorEvidence-driven click-through QA for websites and web applications.

What is inside a skill?

A SKILL.md entry point, focused references, and scripts or tests where deterministic checks help. The registry describes routing and compatibility; CW-AIP defines handoffs between skills.

Skills supply instructions and contracts. They do not supply model subscriptions, CRM accounts, or an autonomous outbound service. A passing validator checks a contract; it does not prove that an AI answer is correct.

OpenAI Marketplace (ChatGPT and Codex)

The repository includes a portable OpenAI plugin manifest at plugin.json and a repository-scoped marketplace at .agents/plugins/marketplace.json. Both point at the canonical skills/ tree; there is no mirrored plugins/ skill tree to drift.

For ChatGPT workspace distribution, import https://github.com/CometWeb-io/agent-skills from Workspace settings → Plugins → Add → Import marketplace. Leave Path empty and leave Branch empty to follow the default branch, or set Branch to main. The marketplace syncs daily after GitHub changes; use Sync now when an immediate refresh is needed. See the full setup guide.

Codex Desktop can discover the same repository-scoped marketplace. The existing per-host installers remain available for local skill directories and are not replaced by the marketplace layer.

Contribute

python3 -m pip install -r requirements-dev.txt
python3 -m pytest
python3 tooling/validate_local.py

See CONTRIBUTING.md, security reporting, and the documentation index. Maintainers can inspect context budgets and eval strength.

MIT license · Changelog