The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions do not contain any patterns attempting to override safety guidelines, bypass system constraints, or extract system prompts. It explicitly defines boundaries for usage, such as prohibiting manipulation or advice in abusive situations.
[DATA_EXFILTRATION]: No evidence of hardcoded credentials, access to sensitive file paths, or network operations to untrusted domains was found. The skill operates within the context of the agent's standard toolset for reading and writing local content.
[REMOTE_CODE_EXECUTION]: The skill does not perform any external downloads of packages or scripts. It is composed entirely of markdown documentation and instructional frameworks.
[COMMAND_EXECUTION]: There are no subprocess calls, shell command executions, or attempts to modify system configurations.
[OBFUSCATION]: The content is presented in clear, readable markdown. No Base64, zero-width characters, or homoglyph-based obfuscation techniques were detected.
[INDIRECT_PROMPT_INJECTION]: The skill has an ingestion surface as it is designed to analyze untrusted text (incoming partner messages).
Ingestion points: Processes messages and context provided by the user in SKILL.md via the 'Message Analysis Framework'.
Boundary markers: Includes 'NOT for' and 'Important Boundaries' sections that explicitly define the scope of the coach.
Capability inventory: Uses standard Read, Write, and Edit tools.
Sanitization: Standard LLM processing is used; no specific sanitization logic is implemented for the analyzed text. As this is a vulnerability surface inherent to the skill's purpose rather than a malicious finding, it is considered safe.

partner-text-coach