The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides explicit instructions to execute shell commands as part of a multi-model fallback cascade.
Evidence: Mentions execution of a Node.js script at <USER_HOME>\\.codex\\.tmp\\marketplaces\\openai-codex\\plugins\\codex\\scripts\\codex-companion.mjs (SKILL.md, Section 3).
Evidence: Instructs the use of codex exec --skip-git-repo-check and claude -p ... for auditing and task sign-off.
[DATA_EXFILTRATION]: The skill accesses local databases and configuration files that store facts, session history, and project-specific conventions.
Evidence: Reads from usmc_memory.db (SQLite), gardener.py, CLAUDE.md, .SYNC, and AGENTS.md to retrieve context and lessons.
Evidence: Mentions persisting state in AUTOMATIONS-MEMORY.md or local buffers.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from observed learning effects and session histories and incorporate them into the agent's reasoning process.
Ingestion points: Data is pulled from the Gardener DB and USMC SQLite DB (SKILL.md, Section 1A).
Boundary markers: Documentation does not specify the use of delimiters or sanitization when processing this external memory data.
Capability inventory: The skill has the ability to execute shell commands and read local files.
Sanitization: No explicit sanitization or validation of the retrieved memory content is mentioned.
[PROMPT_INJECTION]: The skill intentionally uses "injectors" to modify or override the agent's standard reasoning process using tailored patterns.
Evidence: Section 4 "Bereichsspezifische Prompt-Injektion (Tailored Injection Pattern)" describes using a Python script (agy_metacognitive_prompt_injector.py) to inject reasoning strategies.

metacognitive-injectors