The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill references several external resources for project management research and methodology documentation.
The skill links to the DORA 2025 report at dora.dev/research/2025/dora-report/ for software delivery performance insights.
Documentation references the GitHub Spec-Kit at github.com/github/spec-kit for executable specification standards.
The planning methodology incorporates intake from well-known project management platforms such as Linear (linear.app) and ClickUp.
These references are to well-known services and trusted organizations, posing no risk of exfiltration or sensitive data exposure.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection as its core function is to ingest and process external data (Epics, Stories, and task lists).
Ingestion points: Receives task descriptions and epic goals from orchestrator agents (Nexus) or direct user input in SKILL.md.
Boundary markers: While explicit delimiters are not enforced in input, the skill implements a strict 'Step Contract Template' and 'INVEST Checklist' in reference/atomic-step-decomposition.md to validate and scope all output.
Capability inventory: The skill maintains routing capabilities to other agents (Builder, Forge, Scout) and integrates with host-provided task tools for Claude Code (TaskCreate, TaskUpdate) as described in reference/progress-tracking.md.
Sanitization: There is no explicit sanitization of input text, but the skill's instructions strictly prohibit implementation (writing code), limiting the potential impact of an injection to planning-level disruptions.
[COMMAND_EXECUTION]: The skill references standard development tools such as git status, git stash, and git log within its reference/emergency-protocols.md to assist in project stabilization during emergencies. These are used according to standard development practices and do not constitute unsafe execution.

sherpa