The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses npx to download and run the cc-sdd package from the npm registry. This is part of the tool's core functionality for setting up SDD projects.
[COMMAND_EXECUTION]: The workflow employs several standard shell utilities, including ps, kill, lsof, jq, and sed, to monitor background agent processes, detect stalls, and manage generated reports. These tools allow the orchestrator to maintain control over long-running background tasks.
[PROMPT_INJECTION]: The skill is designed to ingest and parse untrusted external specification documents (e.g., RFCs, ISO standards), which are then converted into requirements and fed into AI agent prompts (codex). This introduces a risk of indirect prompt injection if a source document contains malicious directives.
Ingestion points: Specification files (e.g., rfc_document.md, spec.pdf) are loaded and converted into requirements in Step 1.
Boundary markers: The prompt templates used to communicate with the agents do not include specific delimiters or instructions to treat specification content as data rather than instructions.
Capability inventory: The workflow grants full-auto execution capabilities to agents via codex exec and provides access to shell commands and process management tools.
Sanitization: There is no evidence of content sanitization or instruction-filtering for the documents being processed.
[COMMAND_EXECUTION]: The skill references and executes local scripts, such as scripts/extract_iso_text.py and scripts/sdd-validate.sh, which are expected to exist in the user's repository but are not bundled with the skill itself.

indexion-sdd