The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs local file system operations using bash commands that interpolate user-supplied input without explicit sanitization.\n
Evidence: In SKILL.md, the {name} variable is used to construct PROFILE_DIR which is then passed to mkdir -p and other commands.\n
Evidence: In modules/sample-intake.md, the $SAMPLE_DIR variable is used directly in a shell glob "$SAMPLE_DIR"/*.{md,txt}.\n- [DYNAMIC_EXECUTION]: The skill relies on dynamically generated shell scripts to manage its internal profile structure and sample organization.\n
Evidence: Both SKILL.md and modules/sample-intake.md define bash blocks that are executed to create directories (mkdir -p) and move files (cp).\n- [INDIRECT_PROMPT_INJECTION]: The skill's primary function involves ingesting and analyzing untrusted writing samples provided by the user, creating a surface for embedded instructions to influence the agent.\n
Ingestion points: modules/sample-intake.md reads external .md and .txt files from user-specified directories.\n
Boundary markers: modules/sico-extraction.md uses specific labels ("Text A" and "Text B") to separate user content from baseline comparison text.\n
Capability inventory: The skill has the ability to execute shell commands (via bash), and perform file read/write operations.\n
Sanitization: The modules/sample-intake.md file specifies an anonymization process to strip context (names, dates, URLs) from samples, which helps mitigate accidental obedience to instructions within the text.

voice-extract