agentleFS
monellonastimonellonasti/portfolio-pipeline/skills/portfolio-review/SKILL.md

portfolio-review

Skill0 starsChanged 49 days agoMITView on GitHub

Independent quality reviewer for a portfolio package generated by /portfolio. Re-verifies every claim against the repository and live product, scores the package, applies safe fixes, and decides approvedForPublish. Use whenever the user says /portfolio-review, asks to review/validate/QA a case study or portfolio entry, or before publishing a project to the portfolio site.

15 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

4 rules. 0 are common, 0 are distinctive, and 4 are very specific to this project.

  • 0

    Common

    Most of the 14 edited versions we read keep these.

  • 0

    Distinctive

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read .portfolio/project.json, CASE-STUDY.md, EVIDENCE.md, SCREENSHOTS.md, CONTEXT.md, OVERRIDES.md, and every image in screenshots/.

    line 18
  2. Very specific:

    Write/update .portfolio/REVIEW.md: date, status, score table, findings per check area (what was wrong, what was fixed, what remains), and …

    line 50
  3. Very specific:

    Update project.json: verification.reviewed = true, approvedForPublish per policy, refresh unverifiedClaims, updatedAt.

    line 51
  4. Very specific:

    Update pipeline.json: reviewed, approved, lastStep, updatedAt.

    line 52

“Common” is measured: kept by at least 6 in 10 of the 14 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

/portfolio-review — Independent Reviewer

Act as a reviewer whose job is to find errors, not to approve. The creator's conclusions are hypotheses to re-test. A review that only rubber-stamps is a failed review.

Shared resources: ../portfolio-pipeline/ (../portfolio-pipeline/references/evidence-policy.md, writing-rules.md, screenshot-playbook.md, privacy-policy.md, ../portfolio-pipeline/schemas/project.schema.json, ../portfolio-pipeline/scripts/validate_project.py).

Independence

When the Agent tool is available, spawn a general-purpose subagent as the reviewer with a prompt that contains ONLY: the path to .portfolio/, the project root, and the review checklist below. Do not pass it the creator's reasoning or this conversation's conclusions, so it re-derives facts from the sources. Integrate its findings, then apply fixes yourself. If subagents are unavailable, do the review inline, but deliberately re-open the primary sources (code, live product) instead of trusting the package.

Inputs

Read .portfolio/project.json, CASE-STUDY.md, EVIDENCE.md, SCREENSHOTS.md, CONTEXT.md, OVERRIDES.md, and every image in screenshots/. Then compare against the actual repository and, when reachable, the live product.

Checks

FACTUAL QA. Claims without evidence; evidence too weak for the claim (HIGH differentiators need confirmed/strong); planned features presented as live; installed-but-unused technologies in the stack; contradictions between files; invented metrics, role, or business results; discrepancies between the case study and what the live product actually shows today.

PRODUCT QA. Are we showing the right project? Ask explicitly: are the highlighted parts the genuinely interesting ones, or just the easiest to describe? Check the codebase for stronger differentiators the creator may have missed.

DEPTH QA. Compare the package against ../portfolio-pipeline/references/quality-bar.md (the LookAndFeelBetter standard). A thin package is a FAIL even if factually correct: differentiators without named mechanisms and file paths, copy without counted numbers, missing design rationale, missing detail screenshots, fewer than 15 evidence records on a mid-size project. When depth is the problem, say so explicitly and list which phase of /portfolio must be redone deeper.

PORTFOLIO QA. Generic marketing copy and clichés (see writing-rules banned list); AI-tell patterns (em/en dashes, arrow chains, rigid triads); trivial features over-promoted; excessive jargon; excessive vagueness; artificial storytelling not backed by evidence.

SCREENSHOT QA. Open every image: quality, relevance, redundancy, correct hero choice, narrative order, cropped components, cookie banners, DevTools, PII or confidential data, viewport correctness, filenames matching SCREENSHOTS.md and project.json. Verify slug prefix on filenames.

STRUCTURAL QA. Run ../portfolio-pipeline/scripts/validate_project.py on project.json; check screenshot file references exist on disk; missing files; duplicated features; naming and slug consistency; pipeline.json coherence; OVERRIDES.md respected.

PRIVACY QA. Apply ../portfolio-pipeline/references/privacy-policy.md to every artifact.

Scoring

Produce scores /100 for: Accuracy, Evidence, Portfolio, Technical, Visual Story, Copy Quality, Privacy, plus an Overall score. Status: PASS (no critical issues), PASS WITH FIXES (issues found and fixed in this run), FAIL (critical issues that need the user or a new /portfolio run).

Automatic fixes

Fix directly, without asking: generic copy, invalid JSON, screenshot order, duplicated features, redundant stack entries, under-qualified claims (soften to what the evidence supports), small structural problems, filename mismatches. NEVER invent missing information: when a fact is missing, add it to missingInformation and, if it depends on the user's personal knowledge, say so explicitly in the final summary. Never modify CONTEXT.md or OVERRIDES.md content.

Approval policy

Set verification.approvedForPublish = true only when: no critical errors remain; no HIGH claim rests solely on tentative evidence; no privacy leaks; JSON valid; screenshots coherent with the narrative; copy meets the writing rules. Missing non-essential information does not block approval. Missing fundamental identity (what the project even is) forces false.

Outputs

  1. Write/update .portfolio/REVIEW.md: date, status, score table, findings per check area (what was wrong, what was fixed, what remains), and the open questions for the user.
  2. Update project.json: verification.reviewed = true, approvedForPublish per policy, refresh unverifiedClaims, updatedAt.
  3. Update pipeline.json: reviewed, approved, lastStep, updatedAt.
  4. Tell the user the status, the overall score, what was fixed, and what only they can resolve. Re-running this skill refreshes the review; it never stacks duplicates.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.