The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from untrusted sources, creating a surface for potential indirect prompt injection attacks.
Ingestion points: Processes content from Slack, Microsoft Teams, Discord, Jira, GitHub, Notion, Zendesk, and other corporate platforms (SKILL.md).
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are used for external data (SKILL.md).
Capability inventory: Possesses capabilities to write to external services via MCP tools (Slack, Notion) and the GitHub CLI (SKILL.md).
Sanitization: No evidence of sanitization or filtering of external content before processing (SKILL.md).
[DATA_EXPOSURE]: The skill maintains persistent state and sensitive user profiles on the local filesystem.
Sensitive file access: Reads and writes configuration, state, and logs to ~/.claude/state/periscope/, including a distillation of the user's writing style in voice.md (SKILL.md).
[COMMAND_EXECUTION]: The skill utilizes external command-line utilities to interact with surfaces.
Tool usage: Invokes the gh tool to fetch data from GitHub repositories (SKILL.md).

periscope