The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill provides explicit instructions to bypass security controls. It directs the agent to use PowerShell or Bash to perform writes on paths that are normally blocked by the platform's 'mobile-app write-guard hook'.
[COMMAND_EXECUTION]: Instructs the agent to execute shell commands to modify files (using PowerShell here-strings or Bash) and to perform external research via the GitHub CLI (gh search).
[PERSISTENCE]: The skill modifies global configuration files (~/.claude/CLAUDE.md and ~/.claude/GOTCHAS.md) to inject global imports and rules. These modifications persist across sessions and affect the agent's behavior in all future projects.
[INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface where instructions embedded in the 'current conversation context' are harvested and written into durable configuration files.
Ingestion points: Current conversation context (session transcript).
Boundary markers: None identified; the skill relies on the agent to identify 'lessons' without explicit sanitization rules for the harvested text.
Capability inventory: The skill has the capability to write to CLAUDE.md, MEMORY.md, and global GOTCHAS.md using shell commands that bypass standard write guards.
Sanitization: There is no mention of sanitizing or escaping the harvested content before it is interpolated into configuration files.
[EXTERNAL_DOWNLOADS]: Fetches external data from GitHub repositories and web searches to research prior art and verify technical claims during the 'Research & route' stage.

reflect