The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from Architecture Decision Records (ADRs) to generate a control manifest, creating a surface where instructions embedded in those documents could influence the agent.
Ingestion points: ADR files located at docs/architecture/adr-*.md are read and processed in Phase 1 and Phase 2.
Boundary markers: The skill uses md-section.sh to extract specific sections, but it lacks explicit delimiters or instructions for the agent to ignore any embedded directives within the extracted text.
Capability inventory: The skill possesses file writing capabilities (docs/architecture/control-manifest.md) and local bash script execution.
Sanitization: No sanitization or filtering of the markdown content is specified before the text is interpolated into the agent's context.
[DYNAMIC_CONTEXT_INJECTION]: The SKILL.md file uses the !bash syntax to execute a shell command at the moment the skill is loaded.
Evidence: !bash "${CLAUDE_SKILL_DIR}/../../hooks/yaml-helper.sh" resolve_config --keys review_mode,review_gate,automation.
Description: This command dynamically resolves configuration parameters. While providing functionality for environment-aware behavior, it represents a pre-execution command invocation mechanism.
[COMMAND_EXECUTION]: The skill performs its primary functions by invoking several local bash scripts through the Bash tool.
Evidence: The workflow involves executing .claude/scripts/adr-status.sh to list records, .claude/scripts/md-section.sh to extract data, and hooks/yaml-helper.sh for configuration.
Description: These scripts are integral to the skill's operation for file parsing and system reporting within the local project environment.

create-control-manifest