Skills maintenance Audit History
📦

Audit History

maintenance - 13 audits

Version comparison

Capability and finding changes across audited versions, newest first.

VersionDateResultReview itemsChange vs previous
v13 LatestJul 23, 2026, 12:01 PM 1 confirmed0No capability change
v12 Jul 17, 2026, 11:01 AM No confirmed findings0No capability change
v11 Jul 17, 2026, 11:01 AM No confirmed findings0No capability change
v10 Jul 17, 2026, 11:01 AM No confirmed findings0No capability change
v9 Jul 5, 2026, 06:19 AM No confirmed findings0No capability change
v8 Jul 9, 2026, 12:43 PM 2 confirmed0No capability change
v7 Jul 5, 2026, 06:19 AM No confirmed findings0No capability change
v6 Jun 28, 2026, 08:07 PM No confirmed findings3 Contains scripts
v5 Jan 16, 2026, 07:53 PM No confirmed findings0No capability change
v4 Jan 16, 2026, 07:53 PM No confirmed findings0No capability change
v3 Jan 10, 2026, 12:14 PM No confirmed findings0No capability change
v2 Jan 10, 2026, 12:14 PM No confirmed findings0No capability change
v1 Jan 10, 2026, 12:14 PM No confirmed findings0Baseline

Jul 23, 2026, 12:01 PM

All 29 static findings are false positives caused by Markdown formatting, fixed command substitution, intentional project paths, and Japanese text. A semantic review found a medium-severity data-loss risk because deletion and in-place truncation can run without a mandatory preview or confirmation.

2
Files scanned
282
Lines analyzed
3
Review items
0
False positives ignored

Confirmed security concerns (1)

Medium
Destructive Cleanup Without Mandatory Confirmation
The logs workflow uses find -delete, and the state workflow replaces a trace file after truncation. Dry-run is optional, and neither workflow requires explicit confirmation immediately before the destructive action.
The documented commands directly delete old log files and overwrite the trace file. A general backup rule exists, but no confirmation gate or default preview is specified.
Audited by: codex

Jul 17, 2026, 11:01 AM

All 29 static findings are false positives. Most detect Markdown inline-code formatting, while the timestamped backup example uses fixed local shell syntax. The two high-entropy alerts are explained by readable multilingual Markdown; no prompt injection, exfiltration intent, or hidden payload was found.

2
Files scanned
282
Lines analyzed
2
Review items
0
False positives ignored
Audited by: claude

Jul 17, 2026, 11:01 AM

All 29 static findings are false positives. Most detect Markdown inline-code formatting, while the timestamped backup example uses fixed local shell syntax. The two high-entropy alerts are explained by readable multilingual Markdown; no prompt injection, exfiltration intent, or hidden payload was found.

2
Files scanned
282
Lines analyzed
2
Review items
0
False positives ignored
Audited by: claude

Jul 17, 2026, 11:01 AM

All 29 static findings are false positives. Most detect Markdown inline-code formatting, while the timestamped backup example uses fixed local shell syntax. The two high-entropy alerts are explained by readable multilingual Markdown; no prompt injection, exfiltration intent, or hidden payload was found.

2
Files scanned
282
Lines analyzed
2
Review items
0
False positives ignored
Audited by: claude

Jul 5, 2026, 06:19 AM

Most static findings are false positives from Markdown references or documented Bash examples for local cleanup checks. The obfuscation alerts are explained by readable Japanese prose and front matter, and no prompt injection or data exfiltration intent was found.

2
Files scanned
341
Lines analyzed
2
Review items
0
False positives ignored
Audited by: codex

Jul 9, 2026, 12:43 PM

Most static findings are Markdown backtick formatting or false obfuscation heuristics after manual review. The real risk is semantic: the skill guides destructive cleanup of .claude logs and state files, with user-controlled thresholds that need strict confirmation.

2
Files scanned
282
Lines analyzed
4
Review items
0
False positives ignored

Confirmed security concerns (2)

High
Destructive Cleanup Operations
The cleanup reference directs deletion of old .claude logs and truncation of agent state files. These operations can remove audit history or session context.
The file explicitly documents deletion and truncation operations on hidden workspace state. This is intended maintenance behavior, but it is destructive.
Medium
User-Controlled Cleanup Scope
The skill accepts free-form instructions for thresholds, exclusions, and extra archive deletion. A broad prompt can expand the cleanup scope beyond defaults.
The instructions explicitly allow free-form threshold changes and additional archive deletion. The risk depends on confirmation behavior, so confidence is high but not absolute.
Audited by: codex

Jul 5, 2026, 06:19 AM

Most static findings are false positives from Markdown references or documented Bash examples for local cleanup checks. The obfuscation alerts are explained by readable Japanese prose and front matter, and no prompt injection or data exfiltration intent was found.

2
Files scanned
341
Lines analyzed
2
Review items
0
False positives ignored
Audited by: codex

Jun 28, 2026, 08:07 PM

Static analysis reported many command-execution, filesystem, weak-crypto, and entropy findings. Review found no malicious payload, network access, credential handling, or prompt-injection text, but the skill intentionally allows Bash and file editing for project cleanup, so it should publish with a medium-risk warning.

2
Files scanned
341
Lines analyzed
5
Review items
1
False positives ignored
Capability review items (3)

These are real local capabilities that may be expected for this skill, so they require review but are not counted as confirmed malicious behavior.

Medium
Bash-Enabled File Maintenance Workflow
The skill declares Bash as an allowed tool and instructs the agent to read auto-cleanup guidance before acting. The referenced guidance includes shell commands that inspect project files and create archive directories. This is legitimate for cleanup work, but it can alter local project state if used without review.
The Bash permission and shell examples are explicit, and the cleanup purpose is consistent across both files. Confidence is not higher because the commands are documented examples rather than packaged executable scripts.
Medium
Project File Modification and Archiving
The skill is designed to move old tasks, split session logs, create archive directories, and make backups. These actions are expected for maintenance, but accidental data movement or deletion could occur if the agent applies the guidance too broadly.
The file maintenance intent is stated directly, and filesystem changes are part of the documented behavior. The risk remains moderate because the content also mentions backups and preserving in-progress tasks.
Low
Device File Access Alert Is Benign stderr Redirection
The filesystem alert on /dev/null appears in a date parsing fallback that redirects errors. It does not read sensitive device files or persist data.
The cited line uses 2>/dev/null only to suppress date parsing errors. This is a common shell pattern and not suspicious by itself.
Static false positives ignored (1)

These static matches were dismissed by semantic review or matched schema-only tokens, so they are shown for transparency but do not drive the quality score.

Low
Static Weak-Crypto and Entropy Alerts Are False Positives
The high weak-cryptographic-algorithm and entropy alerts point to front matter and Japanese prose, including references to Markdown files. No evidence found of cryptographic code, encoded payloads, binary blobs, or malware-like obfuscation.
Manual review of the cited lines shows YAML front matter and natural-language descriptions. The entropy signal is explained by multilingual text rather than executable or encoded content.

Detected Patterns

External Shell Command ExamplesFilesystem Cleanup Instructions
Audited by: codex

Jan 16, 2026, 07:53 PM

All 60 static findings are FALSE POSITIVES. The flagged code exists in documentation files (auto-cleanup/doc.md), not executable code. Bash command examples use hardcoded strings with no user input. The YAML frontmatter '---' was misidentified as weak cryptography. This is a benign file cleanup utility for managing Plans.md, session-log.md, and CLAUDE.md.

3
Files scanned
592
Lines analyzed
3
Review items
0
False positives ignored
Audited by: claude

Jan 16, 2026, 07:53 PM

All 60 static findings are FALSE POSITIVES. The flagged code exists in documentation files (auto-cleanup/doc.md), not executable code. Bash command examples use hardcoded strings with no user input. The YAML frontmatter '---' was misidentified as weak cryptography. This is a benign file cleanup utility for managing Plans.md, session-log.md, and CLAUDE.md.

3
Files scanned
592
Lines analyzed
3
Review items
0
False positives ignored
Audited by: claude

Jan 10, 2026, 12:14 PM

This is a file cleanup utility skill with minimal risk. It uses bash scripts for standard file operations (read, write, move) on project documentation files. No network access, no credential exposure, and no persistence mechanisms detected. The skill operates within its documented scope of cleaning Plans.md, session-log.md, and CLAUDE.md files.

2
Files scanned
293
Lines analyzed
3
Review items
0
False positives ignored
Audited by: claude

Jan 10, 2026, 12:14 PM

This is a file cleanup utility skill with minimal risk. It uses bash scripts for standard file operations (read, write, move) on project documentation files. No network access, no credential exposure, and no persistence mechanisms detected. The skill operates within its documented scope of cleaning Plans.md, session-log.md, and CLAUDE.md files.

2
Files scanned
293
Lines analyzed
3
Review items
0
False positives ignored
Audited by: claude

Jan 10, 2026, 12:14 PM

This is a file cleanup utility skill with minimal risk. It uses bash scripts for standard file operations (read, write, move) on project documentation files. No network access, no credential exposure, and no persistence mechanisms detected. The skill operates within its documented scope of cleaning Plans.md, session-log.md, and CLAUDE.md files.

2
Files scanned
293
Lines analyzed
3
Review items
0
False positives ignored
Audited by: claude