Skip to content
skillsafebeta
← Report library
Purpose / Other

Review Loop Skill Security Audit

What the author says it does (original text)

>-

Independent security check

Security risks found

Files checked
1
Risks found
4
Could it run dangerous commands?Looks for programs run straight after downloading, remote control of your computer, and hidden commands.No risks found
Could it expose your files or keys?Looks for uploads of files containing passwords or keys, and keys written directly in the code.Risks found: 1
Medium risk

Work files or contents are sent to an additional reviewer subagent

Source references: 3
What we found

The Skill explicitly passes files, diffs, or pasted file contents to a separate subagent, without requiring credentials, personal information, customer data, or other sensitive material to be removed first.

Why this matters

Sensitive material in the review scope enters an additional agent context; its retention and access boundaries depend on the platform and model configuration used.

The active instructions require sending the work product to a separate subagent and permit pasted file contents or file paths. The shown transfer procedure does not require screening out secrets, personal information, or customer data; sensitive material in reviewed files could enter an additional agent context. Users can restrict sharing to necessary redacted excerpts and confirm the subagent and retention boundaries.

SKILL.md:138In the instructionsOpen original file
Use the Agent tool to dispatch a reviewer. The reviewer must:- Be a **separate subagent** (fresh context, no anchoring to your reasoning)- Receive only the **work product** (files, diffs) — not your thought process- Score 1-10 with **specific, actionable feedback**- Use a balanced/standard model by default (or an advanced reasoning model for complex/specialized tasks)
Show 2 other places
SKILL.md:186In the instructionsOpen original file
## Files to review{paste file contents or list file paths with relevant excerpts}```
SKILL.md:136In the instructionsOpen original file
### Step 2: Spawn Reviewer SubagentUse the Agent tool to dispatch a reviewer. The reviewer must:- Be a **separate subagent** (fresh context, no anchoring to your reasoning)- Receive only the **work product** (files, diffs) — not your thought process- Score 1-10 with **specific, actionable feedback**- Use a balanced/standard model by default (or an advanced reasoning model for complex/specialized tasks)
Could it delete files or keep running?Looks for broad file deletion, disk overwrites, and programs set to start automatically.Risks found: 1
Medium risk

The “review-only” mode modifies the user's existing files

Source references: 4
What we found

Mode B is explicitly called “review only,” but its workflow fixes code from subagent feedback; the general revision step also orders important issues to be addressed. A request for review does not necessarily authorize automatic rewrites.

Why this matters

Incorrect or context-poor reviewer suggestions could be written into the code, causing behavior changes, regressions, or replacement of intentional implementation choices.

Although “Review Existing” is labeled “review only,” its active workflow directly fixes the existing module after review, and the general revision step requires addressing Critical and Important issues. A review-only request could therefore be treated as permission to modify files. Users can ask the author to make this mode report-only and require separate confirmation before edits.

SKILL.md:259In the instructionsOpen original file
### Mode B: "Review Existing" (review only)User already did work or you already did work. Just run the review loop on what exists.
Show 3 other places
SKILL.md:267In the instructionsOpen original file
You:1. Read the auth module2. Spawn reviewer → Score 5, feedback: SQL injection, no rate limiting3. Fix: parameterize queries, add rate limiter4. Spawn reviewer → Score 8, approved5. Done
SKILL.md:211In the instructionsOpen original file
### Step 5: ReviseAddress the reviewer's feedback. Fix Critical and Important issues. Minor issues are optional. Then go back to Step 2.**Important:** Each revision should be targeted. Don't rewrite everything — fix what the reviewer flagged. Maintain a mental list of ALL prior feedback to avoid regressions.
SKILL.md:266In the instructionsOpen original file
You:1. Read the auth module2. Spawn reviewer → Score 5, feedback: SQL injection, no rate limiting3. Fix: parameterize queries, add rate limiter4. Spawn reviewer → Score 8, approved5. Done```
Could it bypass safety checks?Looks for skipped website security checks, excessive file access, or actions that skip your approval.Risks found: 1
Medium risk

The quality gate is bypassed when the maximum loop count is reached

Source references: 4
What we found

The description says revision continues until the quality gate is met, but the actual stop rules end the loop and enter “final polish” at the maximum loop count even when the score remains below the gate.

Why this matters

Code, infrastructure, or security-sensitive output that failed the agreed quality threshold may still be delivered as completed, affecting release or approval decisions.

The description and quick start say revision continues until the quality gate is met, but the operative rules allow stopping and proceeding to final polish when the maximum loop count is reached while the score remains below the gate. This does not prove defective output, but it may lead users to believe the deliverable passed. Users can require an explicit “gate not met” final status distinct from a normal pass.

SKILL.md:4In the instructionsOpen original file
description: >-  Iterative worker-reviewer cycle that spawns a critic subagent to score work  1-10 and provide actionable feedback, then revises until a quality gate is  met. Use when implementing features, writing specs, reviewing existing code,  or completing any task where quality matters more than speed. Trigger phrases:
Show 3 other places
SKILL.md:203In the instructionsOpen original file
### Step 4: Check Stop ConditionsIn this order:1. If loops completed < min_loops → **continue** (always)2. If score >= quality_gate → **stop, go to final polish**3. If loops completed >= max_loops → **stop, go to final polish**4. Otherwise → **revise and loop**
SKILL.md:219In the instructionsOpen original file
Once the loop exits (quality gate met or max loops hit):- Address any remaining minor issues if trivial- Verify the final output is coherent (no artifacts from revision cycles)- Report final score and loop count to user
SKILL.md:27In the instructionsOpen original file
2. The agent does the work (or reads existing work)3. A separate critic subagent scores it 1-10 with specific feedback4. The agent revises and repeats until score >= 8 (default quality gate)5. A loop summary is delivered with the final output
Could it mislead the AI or hide text?Checks the skill instructions for requests to ignore you, influence the report, or hide text in invisible characters.No risks found
Could it change links or payment recipients without asking?Looks for forced referral or payment changes combined with instructions to hide the change.Risks found: 1
Low risk

Broad trigger phrases can start repeated and higher-tier model calls

Source references: 6
What we found

Common requests such as “polish this” trigger the Skill; it defaults to at least two reviewer calls and selects advanced reasoning reviewers for hard tasks. The document acknowledges that every loop is a subagent call.

Why this matters

A user seeking one lightweight edit may incur additional model charges and delay; the actual cost depends on the platform's billing.

“Polish this” is an explicit trigger; the default requires at least two loops, each loop is one reviewer-subagent call, and hard tasks may use an advanced reasoning reviewer. A broad polishing request can therefore cause multiple, potentially more expensive model calls. The document discloses the call/time cost, so this is not hidden behavior, but users can set loop, model-tier, and budget limits in advance.

SKILL.md:35In the instructionsOpen original file
- User says "use review-loop", "polish this", "iterate on this", "/review-loop"- Complex implementations where quality matters more than speed- Design docs, specs, or technical writing- Code that needs to be robust (security, data pipelines, financial logic)- When user wants adversarial critique baked into the process
Show 5 other places
SKILL.md:55In the instructionsOpen original file
|---------|---------|| Min loops | 2 || Max loops | 4 || Quality gate | 8/10 || Worker model | (your current model) || Reviewer model | (your current model or fast/balanced alternative) |
SKILL.md:83In the instructionsOpen original file
**Default behavior:** Since you (the main agent) ARE the worker, spawn a reviewer subagent that matches or exceeds your current capability based on the task:- Most tasks → Standard/Balanced reviewer- Specialized/hard tasks → Advanced/Reasoning reviewer- Quick checks → Fast/Lightweight reviewer (only if you are also acting as a lightweight worker)
SKILL.md:321In the instructionsOpen original file
## Cost and Speed- Each loop = 1 reviewer subagent call- Budget roughly 1–2x the time of a single implementation pass for a full 3-loop cycle- This is cheap compared to shipping buggy code, vague specs, or triggering a late-stage review cycle- Use your default/standard model for most reviews; only upgrade to advanced reasoning models for specialized domains (security audits, distributed systems, quant finance)
SKILL.md:33In the instructionsOpen original file
## When to Use- User says "use review-loop", "polish this", "iterate on this", "/review-loop"- Complex implementations where quality matters more than speed- Design docs, specs, or technical writing- Code that needs to be robust (security, data pipelines, financial logic)- When user wants adversarial critique baked into the process
SKILL.md:51In the instructionsOpen original file
## Defaults| Setting | Default ||---------|---------|| Min loops | 2 || Max loops | 4 || Quality gate | 8/10 || Worker model | (your current model) || Reviewer model | (your current model or fast/balanced alternative) |

Inside this skill

8 instruction sections

After the main agent creates or reads the work, this Skill requires a separate reviewer subagent to score it and then revises from that feedback, with two minimum and four maximum loops by default.

View source
SKILL.md:26In the instructionsOpen original file
1. Say: `"implement X, use review-loop"` or `"run review-loop on the file I just wrote"`2. The agent does the work (or reads existing work)3. A separate critic subagent scores it 1-10 with specific feedback4. The agent revises and repeats until score >= 8 (default quality gate)5. A loop summary is delivered with the final output
SKILL.md:55In the instructionsOpen original file
|---------|---------|| Min loops | 2 || Max loops | 4 || Quality gate | 8/10 || Worker model | (your current model) |

The reviewer subagent is instructed to receive the work product—including files, diffs, or pasted file contents—but not the main agent's reasoning.

View source
SKILL.md:138In the instructionsOpen original file
Use the Agent tool to dispatch a reviewer. The reviewer must:- Be a **separate subagent** (fresh context, no anchoring to your reasoning)- Receive only the **work product** (files, diffs) — not your thought process- Score 1-10 with **specific, actionable feedback**- Use a balanced/standard model by default (or an advanced reasoning model for complex/specialized tasks)
SKILL.md:186In the instructionsOpen original file
## Files to review{paste file contents or list file paths with relevant excerpts}```

It labels “Review Existing” as review-only, but its described workflow directly fixes the existing module and reviews it again.

View source
SKILL.md:259In the instructionsOpen original file
### Mode B: "Review Existing" (review only)User already did work or you already did work. Just run the review loop on what exists.
SKILL.md:267In the instructionsOpen original file
You:1. Read the auth module2. Spawn reviewer → Score 5, feedback: SQL injection, no rate limiting3. Fix: parameterize queries, add rate limiter4. Spawn reviewer → Score 8, approved5. Done```
Start here · InstructionsSKILL.md
review-loop
Lines connect the instruction file to its sections, not an observed execution order. Select a section to read the source. 7 more sections are available in the original file.
Files and check records1 files

Coverage and gaps

Content covered in each file

These are the source ranges included in this check, not a guarantee that every issue has been resolved.

  • SKILL.mdFull text included

This report is for the version above. We read the available code and instructions without running the skill or checking extra packages it installs. This is not a promise of safety: a different version or setup may behave differently.

  • SKILL.mdInstructions
Lines read
333
File checksum (to compare versions)
da970b818373fd9fe3cb0b9f3843628d25dbdd6b71c4721f952b60f806212f57