comptable (paperasse) — agentic threat model

Not certain from the listing — the underlying LLM is not specified, but it is vulnerable to prompt injection that could alter accounting logic or trigger malicious parameters in the bundled helper scripts.

Reads bundled data files and references for French accounting. Threat of data poisoning if an attacker modifies the local reference files or inputs malicious financial documents to manipulate the agent's calculations.

Orchestrates tasks by running bundled helper scripts to process documents. Vulnerable to insecure tool integration and local code execution if the framework does not sanitize inputs passed to these scripts.

Not certain from the listing — as an open-source skill pack, deployment depends on the user's local environment. Lack of sandboxing for the helper scripts could allow local privilege escalation or directory traversal.

Not certain from the listing — there are no mentioned guardrails, logging, or evaluation frameworks to detect drift, anomalous script execution, or biased accounting outputs.

Not certain from the listing — no built-in compliance controls, access policies, or audit trails are mentioned for handling sensitive French financial and administrative data.

Part of a 'paperasse' pack of six sibling skills (notaire, fiscaliste, etc.). High risk of cascading failures or trust abuse if these sibling agents interact and pass unvalidated data to one another.