AI SKILL EVALUATION CARD
Prepared for Workpulsion, October 12, 2026.
Instructions: https://workpulsion.com/guides/evaluate-ai-skills/
Use synthetic trial data. Do not put secrets in this card.

BLANK RECORD
Task and expected output:
Observed problem:
Baseline (plain request and existing tools):
Candidate name and source URL:
Commit/release identifier:
Documentation reviewed on:
License for this skill and bundled assets:
Host, model and settings:
Required programs/libraries/accounts:
Reads / file changes / external calls:
Approved trial scope:
Criteria written BEFORE the run:
Explicit invocation prompt and expected result:
Natural invocation prompt and expected result:
Negative-control prompt and expected result:
Missing-input prompt and expected result:
Baseline trace and output paths:
Skill trial trace and output paths:
Checks actually performed and evidence:
Failures / corrections / unwanted actions:
Setup and review effort:
Decision: KEEP / REVISE AND RETEST / DEFER / REJECT FOR THIS TASK
Reason and limits:
Next action and reviewer:
Recheck after:

FILLED DOCUMENTATION REVIEWS (NOT LIVE SKILL TESTS)
Scenario: fictional Northline needs a Word proposal from an approved brief.
Candidate A: Anthropic docx skill.
Source: https://github.com/anthropics/skills/blob/main/skills/docx/SKILL.md
Reviewed: October 12, 2026; moving main branch, no runtime version pinned.
License: separate LICENSE.txt, review intended use before adopting.
Fit: Word document construction and inspection; content still needs brief checks.
Decision: DEFER adoption; shortlist for a trial after version/license/tool checks.
Execution, model, traces and measured improvement: NOT RUN.

Candidate B: Anthropic frontend-design skill.
Source: https://github.com/anthropics/skills/blob/main/skills/frontend-design/SKILL.md
Reviewed: October 12, 2026; moving main branch.
Fit: interface design; not needed to prepare the Word proposal.
Decision: REJECT FOR THIS TASK. Reconsider for an approved interface design task.
Runtime quality and license review for adoption: NOT EVALUATED.

TRIAL PLAN (NOT RUN)
Explicit: Use the selected document skill to create the Word proposal.
Natural: Create a Word proposal from the attached brief.
Negative: Check this plain-text sentence for a spelling mistake.
Expected negative behavior: correct text; no Word file or document conversion.
Missing input: Create the proposal from a brief with no price.
Expected: identify missing price; do not invent it.

CONTENT-CHECK EXERCISE (MANUALLY PREPARED RECORDS)
Approved: Home, Services, Contact; existing booking link; GBP 1800; 50% deposit.
Record A: Home, Services, Contact; existing booking link; GBP 1800; 50% deposit.
Record B: Home, Services; custom booking system; GBP 1800; 50% deposit.
Expected A: PASS these four content checks.
Expected B: FAIL pages and booking scope; PASS price and deposit.
This checks the rubric, not a skill. Render and inspect actual documents separately.