The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructions direct the agent to load sensitive environment variables from a local secrets file and subsequently perform network operations via the AWS CLI.
Evidence: Step 5 in SKILL.md includes the command source ~/.config/zsh/secrets followed by aws s3 cp to upload generated files.
[COMMAND_EXECUTION]: The workflow relies heavily on executing shell commands to generate charts, process site data, and verify network uploads.
Evidence: SKILL.md contains multiple command-line instructions including python3 scripts/make_charts.py, aws s3 cp, curl, and hugo --minify.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from local blog post tables to generate charts, creating an attack surface where malicious data could influence agent behavior.
Ingestion points: The agent is instructed to read model benchmark data from Hugo article files (e.g., content/2026/09/15/novo-llm-benchmark-v4-*).
Boundary markers: Absent; no specific delimiters or instructions to ignore embedded content within the source files are provided.
Capability inventory: The agent has capabilities to execute shell scripts, write to the filesystem, and perform authenticated network uploads via the AWS CLI.
Sanitization: Absent; the provided Python script (scripts/make_charts.py) and the workflow instructions do not include validation or sanitization steps for the data processed from the article tables.

blog-cost-charts