Back to skill

Security audit

verify-conclusion

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rated this skill suspicious because it flagged guidance that could lead to a disruptive service test and API checks using real credentials; the listed files were fully scanned, but Skillspector also reported an exit status of 1, so the results warrant caution.

Consider whether the troubleshooting guidance fits your needs, and be cautious if you would use it on live services or with production credentials. ClawScan’s verdict is suspicious, not malicious. Although the completeness record says all five components were scanned, Skillspector reported exit status 1, so treat the assessment as having some uncertainty.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (14)

The restart-feasibility example calls for a potentially state-changing SIGTERM test despite the read-only verification requirement.

High
Category
Excessive Agency
Confidence
94% confidence
Finding

Checking whether a service responds to SIGTERM can be interpreted as sending the signal, which can terminate a production service and interrupt active work. Although the surrounding instructions require state-preserving checks and explicit approval for changes, this concrete example contradicts that boundary and could lead an agent to treat termination as an authorized diagnostic probe.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · references/pitfalls-01-10.md (reported line 33)May include surrounding context.

md
### ⑤ Intermediate state reported as final conclusion

**Pattern:** A session_search DISCOVERY result shows "X is pending" in the middle of a session, but the same session's bookend_end resolved it. Agent reports the pending state without checking the end.

**Fix:** After any lookup (especially session_search), always check whether the same source later updated or invalidated the finding.

Requires real credentials for API testing without warning about live-service effects or sensitive-data exposure.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

The unconditional instruction to use real credentials when available does not distinguish authorized test credentials from production secrets or constrain the destination and privileges. In an API-diagnostic context, this can turn a harmless mock test into authenticated live traffic with billing, quota, data-access, or credential-exposure consequences, even if the request is otherwise read-only.

Content

No source excerpt is available for this finding.

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · references/pitfalls-11-20.md (reported line 7)May include surrounding context.

md
### ⑪ Output from a restricted tool treated as fact

**Pattern:** The agent verifies whether a file exists using `sudo cat /etc/sudoers.d/X` → sudo refuses because the NOPASSWD scope does not cover `cat` → the `|| echo "FILE NOT FOUND"` fallback fires → the agent asserts "the file does not exist". The file **existed all along**; `sudo` simply lacked permission to run `cat`.

**Root cause:** A false negative caused by a tool-permission limit is taken as truth. `sudo` refusing ≠ the file does not exist — it means "you have no NOPASSWD authorization for this command". Every subsequent step of the diagnosis then rests on a false fact.

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · references/pitfalls-11-20.md (reported line 7)May include surrounding context.

md
### ⑪ Output from a restricted tool treated as fact

**Pattern:** The agent verifies whether a file exists using `sudo cat /etc/sudoers.d/X` → sudo refuses because the NOPASSWD scope does not cover `cat` → the `|| echo "FILE NOT FOUND"` fallback fires → the agent asserts "the file does not exist". The file **existed all along**; `sudo` simply lacked permission to run `cat`.

**Root cause:** A false negative caused by a tool-permission limit is taken as truth. `sudo` refusing ≠ the file does not exist — it means "you have no NOPASSWD authorization for this command". Every subsequent step of the diagnosis then rests on a false fact.

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · references/pitfalls-11-20.md (reported line 11)May include surrounding context.

md
**Root cause:** A false negative caused by a tool-permission limit is taken as truth. `sudo` refusing ≠ the file does not exist — it means "you have no NOPASSWD authorization for this command". Every subsequent step of the diagnosis then rests on a false fact.

**Fix:** Any file-existence check that relies on `sudo` must first be redone with a tool that needs no sudo — `ls -la` or `stat`. File existence (an inode property) and file readability (a permission property) are **two independent propositions**; conflating them produces catastrophic misjudgments.

**Real case (2026-07-27):** Agent asserted `/etc/sudoers.d/<name>` had been deleted → inferred "the Phase 4 fix never ran" → after the user challenged it, re-inspection showed `ls -la /etc/sudoers.d/` listing the file all along, and `sudo -n journalctl --flush` returning 0. The root cause was not a missing file but the permission-limited `sudo cat`.

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · references/pitfalls-11-20.md (reported line 13)May include surrounding context.

md
**Fix:** Any file-existence check that relies on `sudo` must first be redone with a tool that needs no sudo — `ls -la` or `stat`. File existence (an inode property) and file readability (a permission property) are **two independent propositions**; conflating them produces catastrophic misjudgments.

**Real case (2026-07-27):** Agent asserted `/etc/sudoers.d/<name>` had been deleted → inferred "the Phase 4 fix never ran" → after the user challenged it, re-inspection showed `ls -la /etc/sudoers.d/` listing the file all along, and `sudo -n journalctl --flush` returning 0. The root cause was not a missing file but the permission-limited `sudo cat`.

### ⑫ Tool output treated as a logical assertion (the `|| echo` anti-pattern)

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · references/pitfalls-11-20.md (reported line 13)May include surrounding context.

md
**Fix:** Any file-existence check that relies on `sudo` must first be redone with a tool that needs no sudo — `ls -la` or `stat`. File existence (an inode property) and file readability (a permission property) are **two independent propositions**; conflating them produces catastrophic misjudgments.

**Real case (2026-07-27):** Agent asserted `/etc/sudoers.d/<name>` had been deleted → inferred "the Phase 4 fix never ran" → after the user challenged it, re-inspection showed `ls -la /etc/sudoers.d/` listing the file all along, and `sudo -n journalctl --flush` returning 0. The root cause was not a missing file but the permission-limited `sudo cat`.

### ⑫ Tool output treated as a logical assertion (the `|| echo` anti-pattern)

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · references/pitfalls-11-20.md (reported line 28)May include surrounding context.

Print the two separately; do not merge them with ||. Typical wrong form vs correct form:

text
❌ sudo cat /f || echo "FILE NOT FOUND"
❌ curl -s X || echo "API DOWN"
✅ ls -la /f; echo "exit=$?"          ← first confirm the tool itself ran
✅ curl -s -w '\nHTTP:%{http_code}' X ← two lines: body + exit, independent

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · references/pitfalls-11-20.md (reported line 34)May include surrounding context.

✅ curl -s -w '\nHTTP:%{http_code}' X ← two lines: body + exit, independent

text

**Related trap — false independence in Step 2:** when Step 2 uses the same restricted tool as Step 1 (e.g. `sudo test -f` and `sudo cat` share the same NOPASSWD whitelist limit), both "independent" derivations in fact share one hidden premise — a single cause breaks both at once. True independence requires Step 2 to use a **different access path** (e.g. `ls -la` instead of `sudo cat`).

**Same family — the silent-failure aggregator.** Any loop that sums per-object reads with errors discarded undercounts while looking complete:

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · references/pitfalls-11-20.md (reported line 34)May include surrounding context.

✅ curl -s -w '\nHTTP:%{http_code}' X ← two lines: body + exit, independent

text

**Related trap — false independence in Step 2:** when Step 2 uses the same restricted tool as Step 1 (e.g. `sudo test -f` and `sudo cat` share the same NOPASSWD whitelist limit), both "independent" derivations in fact share one hidden premise — a single cause breaks both at once. True independence requires Step 2 to use a **different access path** (e.g. `ls -la` instead of `sudo cat`).

**Same family — the silent-failure aggregator.** Any loop that sums per-object reads with errors discarded undercounts while looking complete:

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · references/pitfalls-11-20.md (reported line 34)May include surrounding context.

✅ curl -s -w '\nHTTP:%{http_code}' X ← two lines: body + exit, independent

text

**Related trap — false independence in Step 2:** when Step 2 uses the same restricted tool as Step 1 (e.g. `sudo test -f` and `sudo cat` share the same NOPASSWD whitelist limit), both "independent" derivations in fact share one hidden premise — a single cause breaks both at once. True independence requires Step 2 to use a **different access path** (e.g. `ls -la` instead of `sudo cat`).

**Same family — the silent-failure aggregator.** Any loop that sums per-object reads with errors discarded undercounts while looking complete:

Ae4

Medium
Category
analysis-evasion
Confidence
80% confidence
Finding

Suspicious Unicode normalization or mixed-script content

Content

No source excerpt is available for this finding.

The mandatory report header requires Chinese labels without offering a language choice.

Low
Category
Output Handling
Confidence
88% confidence
Finding

Every review report must open with a fixed header containing Chinese labels, regardless of the user's language. Unlike the explicitly justified Chinese trigger phrases, this output-language constraint has no documented justification or user opt-in.

Content

No source excerpt is available for this finding.