The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill implements a mechanism to load additional instructions from a path computed at runtime. It checks for an environment variable $POWERSUIT_OVERLAY and instructs the agent to read and follow files within that directory, explicitly stating that the overlay content should take precedence over the skill's own instructions.
Evidence: [ -n "$POWERSUIT_OVERLAY" ] && ls -1 "$POWERSUIT_OVERLAY/sdlc" 2>/dev/null || true and the subsequent instruction: "Where the overlay and this file disagree, follow the overlay."
[COMMAND_EXECUTION]: The skill uses the rsync command to copy template files into the repository. While it targets the plugin's own root directory, it represents an automated file system modification using shell commands.
Evidence: rsync -a --ignore-existing "${CLAUDE_PLUGIN_ROOT}/templates/" ./ in SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The skill identifies "modules" by scanning for specific files and is instructed to read content from the $POWERSUIT_OVERLAY directory. If an attacker can influence these files, they could potentially inject malicious instructions into the agent's context.
Ingestion points: Reads project files (e.g., docs/docs.json, gate/gate.json) and external overlay files ($POWERSUIT_OVERLAY/sdlc/README.md).
Boundary markers: None specified for the ingested content.
Capability inventory: Shell execution (rsync, ls) and file writing (docs/sdlc.md).
Sanitization: The skill does not explicitly mention sanitizing or escaping the content read from these external files before using it to fill docs/sdlc.md or following its instructions.

sdlc-init