The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local repository and external tools to drive its orchestration logic.
Ingestion points: Processes output from session-survey.sh (git state, PRs, Taskwarrior tasks) and the content of docs/blueprint/manifest.json in SKILL.md (Step 1, Step 2, Step 4.3).
Boundary markers: No specific delimiters or "ignore embedded instructions" markers are applied to the ingested data.
Capability inventory: Includes execution of bash commands, Taskwarrior operations, git/gh commands, and invocation of other skills.
Sanitization: The skill uses jq for structure parsing but lacks explicit sanitization of the values (such as task descriptions or PR titles) before displaying them or passing them as arguments to sub-skills.
[COMMAND_EXECUTION]: The skill executes local shell scripts and system utilities to collect environment state.
Evidence: Invokes session-survey.sh and distill-survey.sh located in the shared scripts directory (${CLAUDE_SKILL_DIR}/../../scripts/).
Evidence: Directly executes task, jq, git, and gh via the Bash tool to manage session state and tasks.

session-end