The Agent Skills Directory

[PROMPT_INJECTION]: No attempts to bypass safety guardrails or override system instructions were detected. The skill maintains a narrow focus on text curation logic.
[DATA_EXFILTRATION]: The skill is configured with no network access ('Network: none'). It interacts only with local project files in specific directories (sections/, output/, outline/) and does not access sensitive system paths or credentials.
[REMOTE_CODE_EXECUTION]: There are no patterns indicating the download or execution of remote scripts, binary payloads, or external packages.
[COMMAND_EXECUTION]: The skill does not invoke system-level commands, shell scripts, or subprocesses. Its operations are limited to reading and writing text files.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from draft files (sections/*.md). While this represents an attack surface, the risk is mitigated by the absence of high-privilege capabilities such as network access or code execution that could be exploited by embedded instructions.
[SAFE]: The skill's behavior is consistent with its stated purpose of document synthesis and provides structural guardrails (e.g., citation consistency contracts) to ensure output integrity.

paragraph-curator