Commentary agent skills
This guide covers the published commentary-skills v0.6.2 release, at source commit 6903701f26de9076dc47823adca92037be3edd1e. Its portable skills guide compatible agents through authorized Commentary workflows. A skill is instruction content, not an access grant, API token, human decision, or provider permission.
Install a skill folder
Section titled “Install a skill folder”- Check that your agent harness supports the
SKILL.mdfolder format. - Clone or download the published
v0.6.2tag. For Git, usegit clone --branch v0.6.2 --depth 1 https://github.com/commentary-dev/commentary-skills.git. - Copy the complete selected
skills/<skill-name>/folder to the skill directory documented by your agent. PreserveSKILL.md,references/,agents/, and other files. - Restart or refresh the agent harness if it only discovers skills at startup.
- Ask the agent to use the installed skill by name for a task that matches its purpose.
The skill repository also provides generated plugin bundles for compatible plugin hosts. Use its own installation guide and reviewed release assets for those hosts; a source checkout or ZIP build candidate does not prove a plugin release was published.
Pick a task-specific skill
Section titled “Pick a task-specific skill”The published v0.6.2 release contains the following 21 skills. Each page explains purpose, installation, invocation, workflow, environment, and authority limits.
Durable human-in-the-loop work
Section titled “Durable human-in-the-loop work”- Ask Human
- Request Approval
- Submit Revision
- Escalate Review
- Async Workflow
- Inbox Triage
- Workspace discovery
Review workflows
Section titled “Review workflows”Forms and research
Section titled “Forms and research”- Form creation
- Form rendering
- Form results
- Form result visualizations
- Research workflow
- Effective Forms
- Product research design
- Form design compatibility router
Request and question design
Section titled “Request and question design”Authority and environment
Section titled “Authority and environment”Skills use the CLI, HTTP API, or MCP as appropriate for the installed toolchain. Install and authorize those transports separately. Prefer short-lived, narrowly scoped credentials. Never give an agent authority to fabricate a human Decision, send a message without authorization, create Git commits or PRs, or claim verification without evidence.
There is no published @commentary-dev/agent-sdk npm package. Use HTTP or MCP. No COMMENTARY_TOKEN is included in or stored by a skill. Configure credentials using the transport’s own secure mechanism.