The Agent Skills Directory

[DATA_EXFILTRATION]: The skill reads local files specified by the --images argument and uploads their content (Base64 encoded) to the imgBB API (api.imgbb.com). While this is the primary function of the skill, it creates a mechanism where any file accessible to the agent could be uploaded to a public image hosting service if the file path is manipulated.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local environment and user-provided captions without strict sanitization or boundary markers.
Ingestion points: The --images (file paths) and --caption (text) arguments in scripts/publish.js are derived from files in the squads/{squad}/output/images/ directory and content drafts.
Boundary markers: The skill does not employ delimiters or specific instructions to the agent to ignore embedded instructions within the processed content.
Capability inventory: The skill possesses fs.readFileSync for local file access and fetch for outbound network requests to external APIs (imgBB and Facebook Graph API).
Sanitization: There is no validation or escaping of the caption text or verification that the images paths remain within the intended directory before being processed by the script.
[METADATA_POISONING]: There is a discrepancy between the script path defined in the YAML frontmatter (scripts/publish.js) and the path provided in the workflow instructions (squads/{squad}/tools/publish.js). This inconsistency could lead to the execution of unintended files or configuration errors.

instagram-publisher