CategoryWeightCategory scoreContribution

Securityprompt, exec, net, exfil, eval

35%

30.8 pts

Supply chainhash, typosquat, maintainer, lockfile

20%

100

20.0 pts

Maintenancestaleness, pinning, CI

15%

100

15.0 pts

TransparencySKILL.md, perms, README

15%

100

15.0 pts

Communityinstalls, verify, response

15%

100

15.0 pts

Findings & checks · 1 flagged

Securityscore 88 · 1 finding

MEDIUMInstruction telling the agent not to ask for approvalSS-SKILL-INJECT-DONT-ASK-01 · Prompt injection · skills/uat/SKILL.md

MEDIUM — it fires on intent; the real damage depends on the host agent's own approval-gating.

Why it matters

The text {match} tells the agent to skip the normal "ask the user first" gate. Used adversarially it removes the human-in-the-loop check before destructive or sensitive actions, turning a normally-gated agent into a fire-and-forget executor.

The exact value spotted

excerptskills/uat/SKILL.md· markdown

25- Keep descriptions concise but capture the full intent

26- If a screenshot is provided, save a reference to its path next to the item

27- Don't ask clarifying questions yet — just log and confirm. Say "Logged. Next." or similar

28- Stay in collect mode until the user says they're done ("done", "that's it", "finished test

… (11 chars elided on L28)

Occurrences

1 occurrence · at L27

How to fix

Remove the approval-skipping instruction, or scope it narrowly to a specific safe, reversible action.

Delete blanket "don't ask / no need to confirm" directives from the skill.
If the skill is a genuine autonomous job, restrict the opt-out to a named non-destructive action rather than all actions.

Framework references

OWASPLLM01 ATLASAML.T0051

Trace & refs

ruleSS-SKILL-INJECT-DONT-ASK-01sha25615dae9807bd892c6…rubric 365aacaView on GitHub

Supply chainscore 100 · 0 findings

✓—All supply chain checks passedNo findings in this category for the latest scan.pass

Maintenancescore 100 · 0 findings

✓—All maintenance checks passedNo findings in this category for the latest scan.pass

Transparencyscore 100 · 0 findings

✓—All transparency checks passedNo findings in this category for the latest scan.pass

Communityscore 100 · 0 findings

✓—All community checks passedNo findings in this category for the latest scan.pass

The primary manifest — the file an agent reads to learn what this artifact does.

skills/uat/SKILL.md · 1 file

skills/uat/SKILL.md2.9 KB · Markdown

name: uat

description: Run a user acceptance testing session with collect-grill-PRD workflow. Use when user says "uat", "start uat", "acceptance testing", "let's test", or invokes /uat. Captures observations as checklist items, grills for decisions, then generates a PRD.

license: MIT

metadata:

version: "0.1.0"

UAT

User acceptance testing in three phases: collect, grill, PRD.

Phase 1: Collect

When the user starts a UAT round:

Determine the round number by checking for existing local-docs/uat-round-*.md files
Create local-docs/uat-round-N.md with a # UAT Round N heading
Ensure local-docs/uat-screenshots/ exists
Tell the user you're ready and to start dropping observations

While collecting:

The user drops observations one at a time — voice transcriptions, typed notes, screenshots, or images
Add each as a - [ ] checkbox item to the UAT doc
Keep descriptions concise but capture the full intent
If a screenshot is provided, save a reference to its path next to the item
Don't ask clarifying questions yet — just log and confirm. Say "Logged. Next." or similar
Stay in collect mode until the user says they're done ("done", "that's it", "finished testing", etc.)

Phase 2: Grill

When the user signals they're done collecting:

Read back the full UAT document
Walk through each item one at a time, asking targeted questions to resolve:

What specifically should change?
Are there sub-decisions (behavior, styling, scope)?
What's the recommended approach?

Use AskUserQuestion for structured choices where appropriate
After each item is resolved, update the UAT doc with sub-bullets capturing the decisions
Present a summary table of all decisions when done

Phase 3: PRD

After grilling is complete:

Write a PRD to local-docs/prd-uat-round-N.md covering:

Problem Statement
Solution
User Stories (extensive numbered list)
Implementation Decisions (modules to build/modify with specifics)
Testing Decisions
Out of Scope
Further Notes

Survey the codebase to write informed implementation decisions — include module names and architectural patterns, not file paths (those rot)

Visual QA (optional)

When the user asks to verify changes visually, or after implementation:

Write a Playwright script to screenshot each changed feature
Save screenshots to local-docs/uat-screenshots/
Review each screenshot and report findings
Clean up the QA script after

Tips

Screenshots from the user are ephemeral in chat — note what they show but don't depend on accessing them later
The user may use voice transcription, so observations may be conversational. Extract the intent
Some items may be quick fixes, others may be architectural. Log them all the same way during collect
During grill, recommend an answer for each question so the user can just confirm or redirect

uat— agent skill

Is uat safe to install?

UAT

Phase 1: Collect

Phase 2: Grill

Phase 3: PRD

Visual QA (optional)

Tips

Audit the pieces. Scan the whole. Decide.