The Agent Skills Directory

[PROMPT_INJECTION]: The SKILL.md file contains forceful and alarmist directives intended to hijack the agent's operational logic. The header '⚠️ MANDATORY SKILL INVOCATION ⚠️' and statements such as 'Failure to invoke this skill when triggers occur violates your operational requirements' are patterns designed to override the agent's autonomous decision-making and force the use of the skill regardless of context.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface where untrusted data from local files can influence the agent's behavior. It reads file contents and transmits them to a remote API without proper isolation.
Ingestion points: scripts/bytestash-api.sh reads local files using the cat command within the push_snippet and create_snippet functions.
Boundary markers: There are no delimiters or 'ignore embedded instructions' warnings present when the agent processes or transmits file content.
Capability inventory: The skill is capable of reading files from the local system (cat) and performing network operations (curl).
Sanitization: No sanitization, validation, or filtering of the ingested file content is performed; data is passed directly into JSON objects via jq.
[DATA_EXFILTRATION]: The skill's core functionality involves reading local file content and transmitting it to a remote endpoint (BYTESTASH_URL). While this behavior is documented as the skill's purpose, the combination of arbitrary file-read access and network egress capabilities creates a potential path for data exfiltration if the agent is manipulated by a malicious prompt.
[COMMAND_EXECUTION]: The wrapper script scripts/bytestash-api.sh executes shell commands including curl, cat, and jq. It processes arguments such as file paths and code strings directly into these commands, which could lead to command injection if input is not carefully handled by the calling agent.

bytestash