cellcog

The SKILL.md describes a legitimate multimodal Any-to-Any service (processing many local files and producing many output modalities) which matches the skill name/description. However, the registry metadata claims no required env vars or install, while SKILL.md clearly lists an env var (CELLCOG_API_KEY) and a pip install (cellcog). That discrepancy is an incoherence that should be resolved before trust.

SKILL.md instructs use of explicit absolute local file paths enclosed in <SHOW_FILE> tags and says generated files are "auto-downloaded to your machine" and results are delivered to a session. That means the agent will (if used) send local files to CellCog's cloud for processing; this is expected for the stated purpose but raises data-exfiltration / privacy risk if sensitive files are provided. The instructions do not overreach beyond this purpose, but they give broad discretion to include arbitrary local files.

Installation is via pip (pip: cellcog) as declared in SKILL.md — a moderate-risk, common mechanism. The registry, however, reported 'No install spec' which conflicts with SKILL.md. Because pip packages can contain arbitrary code, verify the package source and inspect it before installing in a production environment.

SKILL.md requests a single API credential (CELLCOG_API_KEY), which is proportionate to an external cloud API. The registry metadata omitted this requirement — a mismatch. No unrelated credentials are requested, which is good; still, you should treat the API key as sensitive and use least-privilege or a test key.

The skill is not force-included (always: false) and does not request system-level persistence or modify other skills. Autonomous invocation is allowed (platform default). There is no evidence it requests elevated system privileges.