The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions do not contain any patterns attempting to override agent behavior or bypass safety guidelines. Instead, the protocol emphasizes data provenance and semantic integrity, specifically instructing the agent to never invent data or mislead the reader.
[DATA_EXPOSURE]: No hardcoded credentials, sensitive file paths, or network exfiltration patterns were detected. The skill focuses on generating SVG artifacts from input data sources within the workspace.
[OBFUSCATION]: A thorough scan for Base64 encoding, zero-width characters, homoglyphs, and hidden text patterns yielded no results. All content is transparent and directly related to data visualization tasks.
[REMOTE_CODE_EXECUTION]: There are no remote scripts, package installations, or piped command executions. The skill.package.json confirms that no external scripts are included, and the scriptPolicy is set to data-only.
[PRIVILEGE_ESCALATION]: No commands for escalating privileges, such as sudo or modification of system files, are present.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface as it ingests external data_source inputs to produce visualizations. However, the instructions include a multi-stage validation protocol (Stages 1 through 8) that requires the agent to verify data provenance and perform a deception audit, which serves as a mitigation against misinterpreting embedded instructions as data.
[DYNAMIC_CONTEXT_INJECTION]: No use of the dynamic execution syntax (exclamation mark followed by backticks) was found in the documentation or instructions.
[METADATA_POISONING]: Metadata fields in SKILL.md and skill.package.json are consistent with the skill's stated purpose and contain no deceptive instructions.

fable-dataviz