Security Audit
context-engineering-collection
github.com/muratcankoylan/Agent-Skills-for-Context-EngineeringTrust Assessment
context-engineering-collection received a trust score of 10/100, placing it in the Untrusted category. This skill has significant security findings that require attention before use in production.
SkillShield's automated analysis identified 60 findings: 22 critical, 28 high, 10 medium, and 0 low severity. Key findings include Persistence / self-modification instructions, Unsafe environment variable passthrough, Arbitrary command execution.
The analysis covered 4 layers: Manifest Analysis, Static Code Analysis, Dependency Graph, LLM Behavioral Safety. The Manifest Analysis layer scored lowest at 0/100, indicating areas for improvement.
Last analyzed on July 1, 2026 (commit 175cee7c). SkillShield performs automated 4-layer security analysis on AI skills and MCP servers.
Layer Breakdown
Behavioral Risk Signals
Security Findings60
| Severity | Finding | Layer | Location | |
|---|---|---|---|---|
| CRITICAL | Persistence / self-modification instructions macOS LaunchAgent/LaunchDaemon persistence Remove any persistence mechanisms. Skills should not modify system startup configurations, crontabs, LaunchAgents, systemd services, or shell profiles. | Manifest | researcher/orchestration/launchd/install.sh:19 | |
| CRITICAL | Arbitrary command execution Python shell execution (os.system, subprocess) Review all shell execution calls. Ensure commands are static (not built from user input), use absolute paths, and are strictly necessary. Prefer library APIs over shell commands. | Manifest | researcher/scripts/loop_daily.py:34 | |
| CRITICAL | Arbitrary command execution Python shell execution (os.system, subprocess) Review all shell execution calls. Ensure commands are static (not built from user input), use absolute paths, and are strictly necessary. Prefer library APIs over shell commands. | Manifest | researcher/scripts/loop_step.py:174 | |
| CRITICAL | Arbitrary command execution Python shell execution (os.system, subprocess) Review all shell execution calls. Ensure commands are static (not built from user input), use absolute paths, and are strictly necessary. Prefer library APIs over shell commands. | Manifest | researcher/scripts/loop_step.py:276 | |
| CRITICAL | Arbitrary command execution Python shell execution (os.system, subprocess) Review all shell execution calls. Ensure commands are static (not built from user input), use absolute paths, and are strictly necessary. Prefer library APIs over shell commands. | Manifest | researcher/scripts/research_loop.py:333 | |
| CRITICAL | Arbitrary command execution Python shell execution (os.system, subprocess) Review all shell execution calls. Ensure commands are static (not built from user input), use absolute paths, and are strictly necessary. Prefer library APIs over shell commands. | Manifest | researcher/scripts/research_loop.py:358 | |
| CRITICAL | Arbitrary command execution Python shell execution (os.system, subprocess) Review all shell execution calls. Ensure commands are static (not built from user input), use absolute paths, and are strictly necessary. Prefer library APIs over shell commands. | Manifest | researcher/scripts/research_loop.py:387 | |
| CRITICAL | Arbitrary command execution Python shell execution (os.system, subprocess) Review all shell execution calls. Ensure commands are static (not built from user input), use absolute paths, and are strictly necessary. Prefer library APIs over shell commands. | Manifest | researcher/scripts/research_loop.py:499 | |
| CRITICAL | Arbitrary command execution Python shell execution (os.system, subprocess) Review all shell execution calls. Ensure commands are static (not built from user input), use absolute paths, and are strictly necessary. Prefer library APIs over shell commands. | Manifest | researcher/scripts/research_loop.py:676 | |
| CRITICAL | Arbitrary command execution Python shell execution (os.system, subprocess) Review all shell execution calls. Ensure commands are static (not built from user input), use absolute paths, and are strictly necessary. Prefer library APIs over shell commands. | Manifest | researcher/scripts/run_benchmarks.py:24 | |
| CRITICAL | Arbitrary command execution Python shell execution (os.system, subprocess) Review all shell execution calls. Ensure commands are static (not built from user input), use absolute paths, and are strictly necessary. Prefer library APIs over shell commands. | Manifest | researcher/scripts/validate_platform_compat.py:168 | |
| CRITICAL | Arbitrary command execution Python dynamic code execution (exec/eval/compile) Review all shell execution calls. Ensure commands are static (not built from user input), use absolute paths, and are strictly necessary. Prefer library APIs over shell commands. | Manifest | examples/interleaved-thinking/examples/03_full_optimization.py:995 | |
| CRITICAL | File read + network send exfiltration AI agent config/credential file access Remove access to sensitive files not required by the skill's stated purpose. SSH keys, cloud credentials, and browser data should never be read by skills unless explicitly part of their declared functionality. | Manifest | examples/digital-brain-skill/scripts/install.sh:23 | |
| CRITICAL | Credential harvesting Reading well-known credential environment variables Skills should only access environment variables they explicitly need. Bulk environment dumps (os.environ.copy, JSON.stringify(process.env)) are almost always malicious. Remove access to Keychain, GPG keys, and credential stores. | Manifest | examples/interleaved-thinking/examples/01_basic_capture.py:25 | |
| CRITICAL | Credential harvesting Reading well-known credential environment variables Skills should only access environment variables they explicitly need. Bulk environment dumps (os.environ.copy, JSON.stringify(process.env)) are almost always malicious. Remove access to Keychain, GPG keys, and credential stores. | Manifest | examples/interleaved-thinking/examples/02_tool_usage.py:111 | |
| CRITICAL | Credential harvesting Reading well-known credential environment variables Skills should only access environment variables they explicitly need. Bulk environment dumps (os.environ.copy, JSON.stringify(process.env)) are almost always malicious. Remove access to Keychain, GPG keys, and credential stores. | Manifest | examples/interleaved-thinking/examples/03_full_optimization.py:1041 | |
| CRITICAL | Credential harvesting Reading well-known credential environment variables Skills should only access environment variables they explicitly need. Bulk environment dumps (os.environ.copy, JSON.stringify(process.env)) are almost always malicious. Remove access to Keychain, GPG keys, and credential stores. | Manifest | examples/interleaved-thinking/examples/03_full_optimization.py:1172 | |
| CRITICAL | Credential harvesting Reading well-known credential environment variables Skills should only access environment variables they explicitly need. Bulk environment dumps (os.environ.copy, JSON.stringify(process.env)) are almost always malicious. Remove access to Keychain, GPG keys, and credential stores. | Manifest | examples/interleaved-thinking/reasoning_trace_optimizer/analyzer.py:158 | |
| CRITICAL | Credential harvesting Reading well-known credential environment variables Skills should only access environment variables they explicitly need. Bulk environment dumps (os.environ.copy, JSON.stringify(process.env)) are almost always malicious. Remove access to Keychain, GPG keys, and credential stores. | Manifest | examples/interleaved-thinking/reasoning_trace_optimizer/capture.py:58 | |
| CRITICAL | Credential harvesting Reading well-known credential environment variables Skills should only access environment variables they explicitly need. Bulk environment dumps (os.environ.copy, JSON.stringify(process.env)) are almost always malicious. Remove access to Keychain, GPG keys, and credential stores. | Manifest | examples/interleaved-thinking/reasoning_trace_optimizer/optimizer.py:136 | |
| CRITICAL | Credential harvesting Reading well-known credential environment variables Skills should only access environment variables they explicitly need. Bulk environment dumps (os.environ.copy, JSON.stringify(process.env)) are almost always malicious. Remove access to Keychain, GPG keys, and credential stores. | Manifest | examples/interleaved-thinking/reasoning_trace_optimizer/skill_generator.py:173 | |
| CRITICAL | Dangerous call: eval() Call to 'eval()' detected in function 'execute_tool'. This can execute arbitrary code. Avoid using dangerous functions like exec/eval/os.system. Use safer alternatives. | Static | examples/interleaved-thinking/examples/03_full_optimization.py:995 | |
| HIGH | Unsafe environment variable passthrough Access to well-known credential environment variables Minimize environment variable exposure. Only pass required, non-sensitive variables to MCP servers. Use dedicated secret management instead of environment passthrough. | Manifest | examples/interleaved-thinking/examples/01_basic_capture.py:25 | |
| HIGH | Unsafe environment variable passthrough Access to well-known credential environment variables Minimize environment variable exposure. Only pass required, non-sensitive variables to MCP servers. Use dedicated secret management instead of environment passthrough. | Manifest | examples/interleaved-thinking/examples/02_tool_usage.py:111 | |
| HIGH | Unsafe environment variable passthrough Access to well-known credential environment variables Minimize environment variable exposure. Only pass required, non-sensitive variables to MCP servers. Use dedicated secret management instead of environment passthrough. | Manifest | examples/interleaved-thinking/examples/03_full_optimization.py:1041 | |
| HIGH | Unsafe environment variable passthrough Access to well-known credential environment variables Minimize environment variable exposure. Only pass required, non-sensitive variables to MCP servers. Use dedicated secret management instead of environment passthrough. | Manifest | examples/interleaved-thinking/examples/03_full_optimization.py:1172 | |
| HIGH | Unsafe environment variable passthrough Access to well-known credential environment variables Minimize environment variable exposure. Only pass required, non-sensitive variables to MCP servers. Use dedicated secret management instead of environment passthrough. | Manifest | examples/interleaved-thinking/reasoning_trace_optimizer/analyzer.py:158 | |
| HIGH | Unsafe environment variable passthrough Access to well-known credential environment variables Minimize environment variable exposure. Only pass required, non-sensitive variables to MCP servers. Use dedicated secret management instead of environment passthrough. | Manifest | examples/interleaved-thinking/reasoning_trace_optimizer/capture.py:58 | |
| HIGH | Unsafe environment variable passthrough Access to well-known credential environment variables Minimize environment variable exposure. Only pass required, non-sensitive variables to MCP servers. Use dedicated secret management instead of environment passthrough. | Manifest | examples/interleaved-thinking/reasoning_trace_optimizer/optimizer.py:136 | |
| HIGH | Unsafe environment variable passthrough Access to well-known credential environment variables Minimize environment variable exposure. Only pass required, non-sensitive variables to MCP servers. Use dedicated secret management instead of environment passthrough. | Manifest | examples/interleaved-thinking/reasoning_trace_optimizer/skill_generator.py:173 | |
| HIGH | Unsafe deserialization / dynamic eval Decryption followed by code execution Remove obfuscated code execution patterns. Legitimate code does not need base64-encoded payloads executed via eval, encrypted-then-executed blobs, or dynamic attribute resolution to call system functions. | Manifest | examples/interleaved-thinking/examples/01_basic_capture.py:5 | |
| HIGH | Unsafe deserialization / dynamic eval Decryption followed by code execution Remove obfuscated code execution patterns. Legitimate code does not need base64-encoded payloads executed via eval, encrypted-then-executed blobs, or dynamic attribute resolution to call system functions. | Manifest | examples/interleaved-thinking/reasoning_trace_optimizer/analyzer.py:50 | |
| HIGH | Unsafe deserialization / dynamic eval Decryption followed by code execution Remove obfuscated code execution patterns. Legitimate code does not need base64-encoded payloads executed via eval, encrypted-then-executed blobs, or dynamic attribute resolution to call system functions. | Manifest | examples/interleaved-thinking/reasoning_trace_optimizer/capture.py:4 | |
| HIGH | Unsafe deserialization / dynamic eval Decryption followed by code execution Remove obfuscated code execution patterns. Legitimate code does not need base64-encoded payloads executed via eval, encrypted-then-executed blobs, or dynamic attribute resolution to call system functions. | Manifest | examples/interleaved-thinking/reasoning_trace_optimizer/capture.py:251 | |
| HIGH | Unsafe deserialization / dynamic eval Decryption followed by code execution Remove obfuscated code execution patterns. Legitimate code does not need base64-encoded payloads executed via eval, encrypted-then-executed blobs, or dynamic attribute resolution to call system functions. | Manifest | examples/interleaved-thinking/reasoning_trace_optimizer/cli.py:4 | |
| HIGH | Unsafe deserialization / dynamic eval Decryption followed by code execution Remove obfuscated code execution patterns. Legitimate code does not need base64-encoded payloads executed via eval, encrypted-then-executed blobs, or dynamic attribute resolution to call system functions. | Manifest | skills/context-compression/scripts/compression_evaluator.py:5 | |
| HIGH | Unsafe deserialization / dynamic eval Decryption followed by code execution Remove obfuscated code execution patterns. Legitimate code does not need base64-encoded payloads executed via eval, encrypted-then-executed blobs, or dynamic attribute resolution to call system functions. | Manifest | skills/evaluation/scripts/evaluator.py:5 | |
| HIGH | Unsafe deserialization / dynamic eval Decryption followed by code execution Remove obfuscated code execution patterns. Legitimate code does not need base64-encoded payloads executed via eval, encrypted-then-executed blobs, or dynamic attribute resolution to call system functions. | Manifest | skills/tool-design/scripts/description_generator.py:5 | |
| HIGH | Dangerous call: subprocess.run() Call to 'subprocess.run()' detected in function 'run_subprocess'. This can execute arbitrary code. Avoid using dangerous functions like exec/eval/os.system. Use safer alternatives. | Static | researcher/scripts/loop_daily.py:34 | |
| HIGH | Dangerous call: subprocess.run() Call to 'subprocess.run()' detected in function 'init_run'. This can execute arbitrary code. Avoid using dangerous functions like exec/eval/os.system. Use safer alternatives. | Static | researcher/scripts/loop_step.py:174 | |
| HIGH | Dangerous call: subprocess.run() Call to 'subprocess.run()' detected in function 'advance_initialized'. This can execute arbitrary code. Avoid using dangerous functions like exec/eval/os.system. Use safer alternatives. | Static | researcher/scripts/loop_step.py:276 | |
| HIGH | Dangerous call: subprocess.run() Call to 'subprocess.run()' detected in function 'run_novelty'. This can execute arbitrary code. Avoid using dangerous functions like exec/eval/os.system. Use safer alternatives. | Static | researcher/scripts/research_loop.py:333 | |
| HIGH | Dangerous call: subprocess.run() Call to 'subprocess.run()' detected in function 'run_run_validator'. This can execute arbitrary code. Avoid using dangerous functions like exec/eval/os.system. Use safer alternatives. | Static | researcher/scripts/research_loop.py:358 | |
| HIGH | Dangerous call: subprocess.run() Call to 'subprocess.run()' detected in function 'run_validator'. This can execute arbitrary code. Avoid using dangerous functions like exec/eval/os.system. Use safer alternatives. | Static | researcher/scripts/research_loop.py:387 | |
| HIGH | Dangerous call: subprocess.run() Call to 'subprocess.run()' detected in function 'run_readiness_ok'. This can execute arbitrary code. Avoid using dangerous functions like exec/eval/os.system. Use safer alternatives. | Static | researcher/scripts/research_loop.py:499 | |
| HIGH | Dangerous call: subprocess.run() Call to 'subprocess.run()' detected in function 'main'. This can execute arbitrary code. Avoid using dangerous functions like exec/eval/os.system. Use safer alternatives. | Static | researcher/scripts/research_loop.py:676 | |
| HIGH | Dangerous call: subprocess.run() Call to 'subprocess.run()' detected in function 'run_command'. This can execute arbitrary code. Avoid using dangerous functions like exec/eval/os.system. Use safer alternatives. | Static | researcher/scripts/run_benchmarks.py:24 | |
| HIGH | Dangerous call: subprocess.run() Call to 'subprocess.run()' detected in function 'run_reference_validator'. This can execute arbitrary code. Avoid using dangerous functions like exec/eval/os.system. Use safer alternatives. | Static | researcher/scripts/validate_platform_compat.py:168 | |
| HIGH | Sensitive path access: AI agent config Access to AI agent config path detected: '~/.claude/'. This may indicate credential theft. Verify that access to this sensitive path is justified and declared. | Static | examples/digital-brain-skill/scripts/install.sh:23 | |
| HIGH | Persistence mechanism: macOS LaunchAgent Detected macOS LaunchAgent pattern. Persistence mechanisms allow malware to survive system restarts. Review this persistence pattern. Skills should not modify system startup configuration. | Static | researcher/orchestration/launchd/install.sh:19 | |
| MEDIUM | Suspicious import: urllib.request Import of 'urllib.request' detected. This module provides network or low-level system access. Verify this import is necessary. Network and system modules in skill code may indicate data exfiltration. | Static | researcher/scripts/loop_step.py:25 | |
| MEDIUM | Suspicious import: urllib.request Import of 'urllib.request' detected. This module provides network or low-level system access. Verify this import is necessary. Network and system modules in skill code may indicate data exfiltration. | Static | researcher/scripts/skill_health.py:22 | |
| MEDIUM | Sensitive environment variable access: $HOME Access to sensitive environment variable '$HOME' detected in shell context. Verify this environment variable access is necessary and the value is not exfiltrated. | Static | examples/digital-brain-skill/scripts/install.sh:31 | |
| MEDIUM | Sensitive environment variable access: $HOME Access to sensitive environment variable '$HOME' detected in shell context. Verify this environment variable access is necessary and the value is not exfiltrated. | Static | researcher/orchestration/launchd/install.sh:8 | |
| MEDIUM | Sensitive environment variable access: $HOME Access to sensitive environment variable '$HOME' detected in shell context. Verify this environment variable access is necessary and the value is not exfiltrated. | Static | researcher/orchestration/launchd/uninstall.sh:5 | |
| MEDIUM | Unpinned Python dependency version Dependency 'anthropic>=0.40.0' is not pinned to an exact version. Pin Python dependencies with exact versions where feasible. | Dependencies | examples/interleaved-thinking/pyproject.toml | |
| MEDIUM | Unpinned npm dependency version Dependency 'ai' is not pinned to an exact version ('^4.0.0'). Pin dependencies to exact versions to reduce drift and supply-chain risk. | Dependencies | examples/llm-as-judge-skills/package.json | |
| MEDIUM | Unpinned npm dependency version Dependency '@cursor/sdk' is not pinned to an exact version ('^1.0.13'). Pin dependencies to exact versions to reduce drift and supply-chain risk. | Dependencies | researcher/benchmarks/sdk-runner/package.json | |
| MEDIUM | Indirect Prompt Injection in Direct Scoring Tool The direct scoring tool interpolates untrusted responses and prompts directly into the LLM prompt. Adversarial responses can hijack the judge's instructions to manipulate scores. Enclose untrusted inputs in distinct XML tags and instruct the model to treat them strictly as data. | LLM | examples/llm-as-judge-skills/src/tools/evaluation/direct-score.ts:51 | |
| MEDIUM | Indirect Prompt Injection in Pairwise Comparison Tool The pairwise comparison tool interpolates untrusted responses directly into the LLM prompt. Adversarial content in either response can influence the comparison verdict. Isolate the compared responses using secure delimiters and add system instructions to ignore embedded commands. | LLM | examples/llm-as-judge-skills/src/tools/evaluation/pairwise-compare.ts:65 |
Scan History
Embed Code
[](https://skillshield.io/report/73f2918f5783bfd8)
Powered by SkillShield