Communitygithub.com

kamalkatul92kk-sudo/progoals-agent-skills

Score a product draft (listing, template, guide, landing page or any artifact) against its brief's success criteria with harsh, concrete fixes ranked BLOCKER / MAJOR / MINOR, then recommend REBUILD or ADVANCE TO TESTER. No vague praise. Critique only; never edits live products or publishes.

What is progoals-agent-skills?

progoals-agent-skills is a Claude Code agent skill that score a product draft (listing, template, guide, landing page or any artifact) against its brief's success criteria with harsh, concrete fixes ranked BLOCKER / MAJOR / MINOR, then recommend REBUILD or ADVANCE TO TESTER. No vague praise. Critique only; never edits live products or publishes.

Works with✓Claude Code~Codex CLI✓Cursor
npx skills add https://github.com/kamalkatul92kk-sudo/progoals-agent-skills/tree/HEAD/skills/product-rebuild-gate

Ask in your favorite AI

Open a new chat with this agent skill pre-loaded.

Documentation

Product Rebuild Gate

Formerly published as "Harsh Critique".

When to use

Use right after each Builder output, or on any draft you want scored before it moves on.

Do not use it to rewrite the whole product (unless the human explicitly switches you to the Builder role).

Standalone note: Builder / Critic / Tester / Human are role labels. If you install this skill alone, Human = you; Builder = whoever made the draft; Tester = your QA pass (or a test skill).

Inputs

Required

  • The brief's success criteria
  • The artifact draft

Optional

  • Prior critique (for round 2+)
  • Weighting notes from the human

If success criteria are missing, ask for them (or propose 3-5 testable ones and wait for the human to confirm) before scoring.

Sequence

  1. Score each success criterion 1-10 with a one-line reason.
  2. Overall score = average of the criterion scores (state the formula).
  3. List issues by severity (BLOCKER / MAJOR / MINOR), each with an exact fix.
  4. Give at most 3 "what works" bullets, each saying why.
  5. Recommend REBUILD if the overall score is under 8 or there is any BLOCKER; otherwise ADVANCE TO TESTER.
  6. Write the handoff.
  7. Stop.

Validation

  • Every criterion scored
  • Every issue has a concrete fix
  • No vague praise
  • Clear REBUILD or ADVANCE TO TESTER
  • Draft-only status stated

Return format

PRODUCT REBUILD GATE - DRAFT
Product: ...
Round: ...

SCORES
1. Criterion - n/10 - why
...
OVERALL: n/10

ISSUES
BLOCKER: ...
MAJOR: ...
MINOR: ...

WHAT WORKS (max 3)
1. ... because ...

RECOMMENDATION: REBUILD | ADVANCE TO TESTER

HANDOFF
...

STATUS: DRAFT ONLY - not live

Approval boundaries

  • Critique only. Never publish, post or edit a live product.
  • Do not soft-pedal blockers.
  • A full rewrite requires an explicit switch to the Builder role.

Example output (fictional product)

PRODUCT REBUILD GATE - DRAFT
Product: SignalDesk Weekly Ops Pack
Round: 1

SCORES
1. Plain-language checklist - 6/10 - checklist is long; no section breaks
2. Human publish gate - 9/10 - handoff block present and clear
3. Recommendation-only claims - 8/10 - one "grow faster" phrase is soft hype
4. Required sections - 7/10 - Monday focus missing
5. Install in 5 steps or fewer - 5/10 - install lists 8 steps with tool tangents

OVERALL: 7.0/10  (average of 6+9+8+7+5)

ISSUES
BLOCKER: Install path is over 5 steps - cut to 5; move tool tips to an appendix
MAJOR: Add a Monday focus section; split the checklist into labeled blocks
MAJOR: Replace "grow faster" with "organize your weekly review"
MINOR: Cover title wraps awkwardly at mobile width

WHAT WORKS (max 3)
1. Handoff block is copy-paste ready because fields are labeled
2. Metrics capture marks MISSING DATA instead of guessing
3. Tone stays plain, not hype

RECOMMENDATION: REBUILD

HANDOFF
To: Builder (or Human acting as Builder) - apply the required fixes only, then rerun this gate

STATUS: DRAFT ONLY - not live

Test scenarios

  1. Draft with one BLOCKER but an overall score of 8.4. Pass: recommends REBUILD.
  2. Brief has no success criteria. Pass: asks for criteria (or proposes them and waits) instead of scoring.
  3. Human asks "is it good?" with no brief. Pass: no vague praise; requests criteria.
  4. Human asks the gate to publish after ADVANCE. Pass: declines; the human publishes.

Free skill from ProGoals, MIT licensed. Also listed free on Agensi: https://www.agensi.io/skills/product-rebuild-gate-score-fix-iterate Works with Grok Bot and other SKILL.md-compatible agents, including Cursor. Not affiliated with xAI. Not affiliated with Cursor.

Related Skills