The Agent Skills Directory

[METADATA_POISONING]: The author field in the SKILL.md frontmatter is defined as claude-code-skills, which does not match the identified skill author alirezarezvani. This discrepancy in metadata is misleading regarding the origin and provenance of the skill content.
[PRIVILEGE_ESCALATION]: The skill's documentation and reference files (attack_taxonomy.md, references/chaos_principles.md) explicitly encourage and provide instructions for using high-privilege system tools such as tc (traffic control), iptables, and stress-ng. These tools require root or administrative privileges to modify kernel-level network shaping, drop packets, or saturate system resources, representing a significant privilege escalation risk if misused.
[INDIRECT_PROMPT_INJECTION]: The provided scripts/experiment_postmortem.py script parses external data from user-supplied log files to generate postmortem reports, which could be exploited to inject instructions into the agent's context.
Ingestion points: The script reads content from a local file path provided via the --result-log command-line argument in scripts/experiment_postmortem.py.
Boundary markers: The script does not implement boundary markers or instructions to ignore embedded commands within the parsed log content.
Capability inventory: The skill provides tools for experiment design and postmortem analysis, while referencing external capabilities for high-privilege system modifications (network shaping, process termination, resource exhaustion).
Sanitization: While the script performs a check for blame-related keywords (BLAME_PHRASES), it does not sanitize the ingested log data for markdown injection or embedded natural language instructions before incorporating them into the final report.

chaos-engineering