The Agent Skills Directory

[NO_CODE]: The skill package is composed entirely of Markdown instructions, YAML metadata, and JSON evaluation cases. It does not include any Python scripts, JavaScript files, binaries, or other executable components.
[COMMAND_EXECUTION]: There are no commands executed by the skill's instructions. The README mentions standard installation commands using the 'skills' CLI (e.g., npx skills add xerrion/agent-skills), which are documented for installation purposes rather than runtime execution.
[PROMPT_INJECTION]: The instructions do not attempt to bypass safety filters or override system prompts. Instead, they explicitly reinforce existing boundaries, stating that the skill 'does not authorize edits, commands, network access' and that the agent must 'preserve higher-priority instructions' and 'approved scope'.
[DATA_EXFILTRATION]: No network operations or requests to non-whitelisted domains are present. The skill logic focuses on analyzing information already available in the agent's context and explicitly warns against inventing or searching for external data without authorization.
[INDIRECT_PROMPT_INJECTION]: While the skill is designed to process untrusted user data (technical specifications, logs), it includes robust instructions to mitigate injection risks. It requires the agent to 'separate evidence from assumptions', 'never invent a citation', and mark missing information clearly rather than obeying embedded commands in the processed data.
[SAFE]: References to external sites (agentskills.io, opencode.ai) and the author's GitHub repository (xerrion/agent-skills) are legitimate documentation links and community resources. No obfuscation or malicious redirection was detected.

first-principles