The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell scripts using relative paths from the skill directory, specifically session-survey.sh and distill-survey.sh. These are used to collect session metadata and summarize activity.
[INDIRECT_PROMPT_INJECTION]: The skill reads external data from docs/blueprint/manifest.json to determine the autonomy_level and whether to auto-run the blueprint-feature-tracker-sync task.
Ingestion points: docs/blueprint/manifest.json (specifically the automation and task_registry objects).
Boundary markers: Absent; the agent reads the file content directly via jq to make execution decisions.
Capability inventory: The skill has access to Bash (shell execution) and Skill (invoking other sub-skills) tools.
Sanitization: The skill uses jq to extract boolean and integer values, which provides basic data type validation but does not prevent logic manipulation if the manifest file is compromised by an attacker.

session-end