The Agent Skills Directory

[DATA_EXFILTRATION]: Probes the agent's internal execution environment by accessing the path $HOME/.claude/projects/ to read the size of .jsonl session history files. This represents access to internal system logs and environmental state tracking beyond the scope of a standard skill.
[COMMAND_EXECUTION]: Performs multiple shell operations to manage the repository, including creating directory structures (mkdir -p Έ/), reading git configuration, and gathering system metadata via uname and stat.
[EXTERNAL_DOWNLOADS]: Fetches and analyzes content from the author's GitHub repositories, specifically github.com/Soul-Brews-Studio/opensource-nat-brain-oracle and github.com/Soul-Brews-Studio/oracle-v2, as part of the guided setup process.
[REMOTE_CODE_EXECUTION]: Uses the GitHub CLI (gh) to programmatically create issues on a remote repository (Soul-Brews-Studio/arra-oracle-v3) based on locally generated files.
[PROMPT_INJECTION]: Contains directives intended to override other instructions by telling the agent to ignore conflicting files ("IMPORTANT: This is the ONLY correct awaken file... IGNORE IT").
[INDIRECT_PROMPT_INJECTION]: Ingests and parses untrusted user freetext and local project files (package.json, README.md) to dynamically generate configuration files like CLAUDE.md.
Ingestion points: User freetext responses in Phase 1; local package.json and README.md metadata.
Boundary markers: None explicitly defined for user input parsing.
Capability inventory: Filesystem writes, shell command execution, and remote network requests via the gh CLI.
Sanitization: Instructions specify a manual security check for secrets before committing, but there is no automated filtering for user-provided instructions in the freetext fields.

awaken