The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses git and gh (GitHub CLI) to automate repository workflows. In Phase 5.3, it uses gh pr merge --auto to automatically merge research reports into the primary branch, granting the agent significant autonomy over the codebase without requiring human approval.
[COMMAND_EXECUTION]: Shell instructions in Phase 0.2 and Phase 5.2 interpolate variables extracted from pulse-config.yaml directly into shell commands. While word splitting limits direct command execution, processing unsanitized configuration data in shell loops represents a risk if the environment is compromised.
[PROMPT_INJECTION]: The skill retrieves and processes untrusted web content via WebSearch and WebFetch in Phase 2, which is then summarized and committed to a repository. This makes the skill susceptible to indirect prompt injection.
Ingestion points: External web pages and YouTube content fetched during the research phase (SKILL.md, Phase 2).
Boundary markers: Absent. The skill does not provide instructions to sub-agents to ignore or delimit instructions found within retrieved web data.
Capability inventory: File system writing (Phase 4), Git branch creation, committing, and automated Pull Request management (Phase 5).
Sanitization: Absent. External content is summarized and integrated directly into the final research report and PR body.
[EXTERNAL_DOWNLOADS]: The agent uses web search and fetch tools to retrieve content from third-party domains specified in the user-configured research-sources.yaml.

daily-research