The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions mandate the execution of a local shell script as a validation step. Evidence: SKILL.md contains the instruction to 'Run scripts/graph-audit on every touched durable file' during the mandatory filing gate.- [DYNAMIC_EXECUTION]: The skill uses a subprocess call to run an external script whose content is not defined within the skill itself. Evidence: SKILL.md Step 4 requires the agent to execute scripts/graph-audit and fix violations based on the exit code.- [DATA_EXFILTRATION]: The skill uses directory traversal to read files located outside the skill's base directory. Evidence: SKILL.md references ${CLAUDE_SKILL_DIR}/../../CONVENTIONS.md as the source for shared logic and workflow rules.- [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted external content such as transcripts and meeting notes which are then stored in durable project files. 1. Ingestion points: SKILL.md identifies 'pastes a transcript or meeting notes' as a trigger for the skill. 2. Boundary markers: The skill does not define specific delimiters or security wrappers for the pasted content. 3. Capability inventory: The skill has the capability to write to multiple wiki files and execute a local validation script. 4. Sanitization: No sanitization or escaping steps are defined for the verbatim extractions described in references/debrief-extraction.md.

bank