The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill clones the PAI repository from a public source to facilitate upgrades.
Evidence: git clone https://github.com/danielmiessler/PAI.git in SKILL.md.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute system commands for repository management and file comparison.
Evidence: Usage of git clone, git checkout, and diff -r within the SKILL.md process steps.
[INDIRECT_PROMPT_INJECTION]: The skill processes external repository content which is then integrated into the agent's local configuration, presenting an attack surface if the upstream source is compromised.
Ingestion points: SKILL.md (Phase 1: git clone).
Boundary markers: Absent; the skill relies on manual/agent classification logic rather than technical delimiters.
Capability inventory: The skill uses Read, Write, Edit, and Bash to modify files in ~/.claude/system/, ~/.claude/skills/, and ~/.claude/hooks/ (SKILL.md Phase 5).
Sanitization: Absent; the skill lacks explicit sanitization or filtering of the ingested external content before it is applied to the local system.

pai-upgrade