The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local shell script .specify/scripts/bash/adr-context.sh using $ARGUMENTS as parameters in Step 2. Because the arguments are interpolated directly into the command line without shell-escaping or validation, a user could provide shell metacharacters (e.g., ;, &, |, `) to execute arbitrary commands.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from spec.md and remote Architecture Decision Records (ADRs) from the dotCMS/platform-adrs repository.
Ingestion points: Data is read from .specify/feature.json, the local spec.md file, and remote Markdown files via the gh api tool.
Boundary markers: The instructions do not define boundary markers or provide explicit warnings to the agent to disregard instructions embedded within the ingested ADR content.
Capability inventory: The skill has the capability to execute shell commands and utilize the GitHub CLI tool.
Sanitization: There is no evidence of content sanitization or filtering applied to the retrieved text before it is summarized for the plan.
[COMMAND_EXECUTION]: In Step 3, the skill uses the GitHub CLI (gh api) to fetch file contents. While the repository target is fixed to dotCMS/platform-adrs, the specific <file> parameter is derived from the output of a previous script, which could lead to path traversal or unintended API access if the search output is manipulated.

speckit-adr-context