CtrlK
BlogDocsLog inGet started
Tessl Logo

igmarin/agnostic-planning-skills

Language-agnostic AI knowledge registry for Technical Project Management, PRDs, PRD review, Software Architecture planning, Task breakdown, Estimation, Risk assessment, Status reporting, Backlog prioritization, Sprint planning, Retrospectives, and Agile ticket generation. Uses Markdown + Front-matter architecture.

95

1.03x
Quality

97%

Does it follow best practices?

Impact

95%

1.03x

Average score across 10 eval scenarios

SecuritybySnyk

Passed

No known issues

Overview
Quality
Evals
Security
Files

SKILL.mdskills/prd/review-prd/

name:
review-prd
license:
MIT
description:
Reviews a PRD for completeness testability clarity feasibility scope dependencies and edge cases — classify each finding as Critical Suggestion or Note, cite the specific PRD section or line as evidence but redact sensitive data (API keys tokens passwords), produce a findings table with severity and recommendation, flag ambiguous modal verbs like "should" vs "must". Language-agnostic — evaluates structure and content, not technology. Trigger words: review PRD, PRD review, validate PRD, feasibility check, product requirements document review, PRD checklist.
metadata:
{"version":"1.0.0","user-invocable":"true"}

Reviewing a Product Requirements Document

Evaluate a PRD for quality — not agreement with a preferred solution. Focus on whether it's complete, clear, and actionable.

Quick Reference

  • Input: A PRD (from create-prd or existing document).
  • Output: Findings as Critical, Suggestion, or Note.
  • Checks: Completeness, testability, ambiguity, feasibility, edge cases, dependencies.
  • Rule: Every finding cites the specific PRD section as evidence; redact sensitive data rather than reproducing it verbatim.

HARD-GATE

DO NOT review the idea — review the document's quality.
DO NOT suggest alternative solutions unless a requirement is infeasible.
EVERY finding MUST cite the specific PRD section, line, or requirement — if the cited text contains sensitive data (API keys, tokens, passwords, credentials), abstract or redact it rather than reproducing it verbatim.

Core Process

  1. Receive the PRD — from create-prd, user document, or /tasks/prd-*.md.
  2. Scan the checklist (see Review Checklist below) — covers completeness, testability, clarity, feasibility, scope, dependencies, and edge cases.
  3. Classify: Critical (blocks implementation), Suggestion (improves it), Note (observation).
  4. Produce — findings table with severity, evidence, and recommendation.
  5. Verdict: Approved / Approved with Suggestions / Needs Revision.
  6. Feedback loop: If verdict is Needs Revision, instruct the author to address Critical findings and resubmit. Re-run this review on the updated PRD before proceeding downstream.

Review Checklist

Apply all applicable items. Skip items that are genuinely not relevant to the PRD's domain.

Completeness

  • Goals and success metrics are stated and measurable.
  • All in-scope features are listed; out-of-scope is explicit.
  • Actors / user roles are identified.
  • Non-functional requirements (performance, security, accessibility) are present.

Testability

  • Each requirement can be verified with a concrete pass/fail test.
  • Acceptance criteria are specific (no vague words like "fast", "easy", "good").
  • Edge cases and error states are addressed.

Clarity

  • No undefined acronyms or jargon without a glossary entry.
  • No ambiguous modal verbs ("should" vs. "must" vs. "may").
  • Contradictions between sections are absent.

Feasibility

  • Timeline and milestones are realistic given stated scope.
  • Dependencies (systems, teams, data) are identified.
  • Technical constraints or assumptions are explicit.

Scope & Edge Cases

  • Rollback / failure scenarios are described.
  • Data migration or compatibility concerns are addressed if applicable.
  • Regulatory or compliance requirements are noted if applicable.

Output Style

  1. Verdict — Approved / Approved with Suggestions / Needs Revision.
  2. Findings table| # | Severity | Section | Finding | Evidence | Recommendation |
  3. Summary — count by severity + one-paragraph assessment.
  4. What's Good — acknowledge well-written sections.
  5. English only unless user requests otherwise.

Example Findings Table

#SeveritySectionFindingEvidenceRecommendation
1Critical§3 Acceptance Criteria"The page must load quickly" is not testable — no threshold defined.§3: "The dashboard shall load quickly under normal conditions."Replace with a measurable SLA, e.g. "Dashboard initial load ≤ 2 s at p95 on a 10 Mbps connection."
2Suggestion§5 DependenciesExternal payments API dependency is mentioned but no fallback behavior is specified.§5: "Integration with PaymentsProvider v2 API."Add a requirement describing user-facing behavior when the API is unavailable (timeout, retry, graceful error).
3Note§1 GoalsSuccess metric for user adoption is aspirational but not tied to a measurement method.§1: "Achieve high user adoption within 90 days."Consider specifying the data source (e.g. analytics event) used to measure adoption.

Integration

SkillWhen to chain
create-prdReview immediately after PRD generation
generate-tasksAfter review passes, proceed to task breakdown
tech-lead agentFor deeper feasibility and estimation quality review

skills

CODE_OF_CONDUCT.md

CONTRIBUTING.md

README.md

tile.json