The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill collects user descriptions to derive a change name, which is then used as an argument in shell commands. This represents a potential vulnerability surface if malicious input is provided by a user.
Ingestion points: User input for the change description or name is collected in Step 1 of SKILL.md using the AskUserQuestion tool.
Boundary markers: The skill instructions do not specify explicit delimiters (like triple quotes or XML tags) for user input, though it does include validation rules.
Capability inventory: The skill executes shell commands via the Bash tool, including openspec new change, openspec status, and openspec instructions.
Sanitization: The skill includes a guardrail requiring the agent to derive a kebab-case name and to ask the user for a valid name if the format is incorrect, which serves as a basic sanitization step.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to run the openspec CLI. While the skill's scope is restricted to the openspec prefix in the allowed-tools frontmatter, it still performs active command execution on the host system to manage specifications and artifacts.

openspec-new-change