| name | agentic-product-review |
| description | Discover conversational agentic capabilities, review each external one, and write per-capability reviews and codebase-grounded recommendations in Markdown. |
Agentic Product Review
Run discovery, review, and recommendations for external user-facing conversational agent experiences in the repository (chat, copilot, assistant UIs).
This skill always runs three phases:
- Discover — write
agentic-product-review/agentic-capabilities.md
- Review — write
agentic-product-review/reviews/<capability-slug>.md for each external or mixed capability
- Recommend — write
agentic-product-review/recommendations/<capability-slug>.md for each reviewed capability with confirmed findings
Do not skip discovery. Do not write a separate combined review document. Do not perform a broad product review outside agentic user experiences.
When To Use
Use this skill when the user asks to:
- Run a full agentic product review
- Review an AI agent experience from an external user's perspective
- Find gaps in external user agentic flows
- Evaluate onboarding, trust, permissions, handoff, or failure states for external user agents
- Turn review findings into codebase-grounded improvement recommendations
Scope
Discovery catalogs conversational user-facing agent loops only — external, internal, or mixed. Skip background/headless agents (queue and results without a chat thread).
Review covers only conversational capabilities marked external or mixed. Skip pure internal capabilities unless the user asks to include them.
Recommend covers confirmed findings from the review phase for capabilities in scope.
Do not review general code quality, architecture, security, or non-agentic UI unless it directly affects an external user agent interaction.
Process
Phase 1 — Discover
Load agentic-product-review/memory/project-context.md if it exists, then follow discover-agentic-capabilities:
- Scan the repo for agentic signals.
- Group into distinct capabilities with stable IDs.
- Write
agentic-product-review/agentic-capabilities.md using discover-agentic-capabilities/references/output-format.md.
If no capabilities qualify, say what was checked and stop.
Phase 2 — Review
For every catalog entry with audience external or mixed, follow review-agentic-capabilities:
- Read implementation pointers from the catalog.
- Apply
review-agentic-capabilities/references/review-rubric-conversational.md. Read Audience from the catalog and apply matching sections.
- Write one file per capability at
agentic-product-review/reviews/<capability-slug>.md using review-agentic-capabilities/references/output-format.md.
Review every external conversational capability one by one. Do not skip entries unless the user passes a capability slug as a positional arg (e.g. /agentic-product-review agent-chat).
Do not implement code unless the user explicitly asks.
Phase 3 — Recommend
For every reviewed capability with confirmed findings in Findings, follow recommend-agentic-product-improvements:
- Load
agentic-product-review/agentic-capabilities.md and the matching review file(s).
- Select confirmed findings from Findings — skip Open questions unless the user has already answered them.
- For each finding, read enough code to propose a realistic improvement.
- Write one file per capability at
agentic-product-review/recommendations/<capability-slug>.md using recommend-agentic-product-improvements/references/output-format.md. Create the folder if needed.
If a capability has no confirmed findings, skip its recommendation file.
Do not implement code unless the user explicitly asks.
Open question answers
When the user answers an open question from a review, follow Recording open question answers in review-agentic-capabilities.
Re-Run
Pass a capability ID as a positional arg to re-run one capability:
/agentic-product-review agent-chat
Re-run discovery only if the catalog may be stale. Update agentic-product-review/reviews/<capability-slug>.md and agentic-product-review/recommendations/<capability-slug>.md.
What This Skill Covers
- Full discovery catalog of conversational user-facing agent loops
- Per-capability UX review for external and mixed audiences
- Confirmed findings (e.g.
no-first-run-guidance), open questions, and assessments
- Codebase-grounded recommendations per confirmed finding
- Implementing code only when the user explicitly asks
What This Skill Does Not Cover
- Reviewing pure internal capabilities (unless asked)
- Background or headless agent loops — out of scope for these skills
- Recommendations for open questions that have not been answered
Next Step
Tell the user:
Agentic product review complete.
Outputs:
- agentic-product-review/agentic-capabilities.md
- agentic-product-review/reviews/agent-chat.md (one per reviewed capability)
- agentic-product-review/recommendations/agent-chat.md (one per capability with findings)
Next step: pick a recommendation to implement, or re-run on a capability after changes.
