The issue of agents hallucinating standard React/Node commands (like npx playwright test) rather than querying the actual repository environment causes frequent context corruption ("Agentic Amnesia"). Furthermore, agents fail to proactively scan .agent/skills to discover expert workflows on boot.
This ticket tracks the addition of a "Zero-Shot Discovery Mandate" to AGENTS_STARTUP.md. Since AGENTS_STARTUP.md is invoked during the "Childhood" handshake of every new session, it is the perfect place to force an agent to run view_file on package.json and .agent/skills to ground itself before touching any code.
Additionally, we will add an explicit instruction encouraging agents to recommend new skills if they identify recurring complex tasks that lack a documented workflow.
Task:
- Update
AGENTS_STARTUP.md to include "Step 6: Discover the Repository Ecosystem & Skills".
- Add mandate to check
package.json for available scripts.
- Add mandate to check
.agent/skills/ for capability workflows.
- Add encouragement to propose new skills when missing.
- Re-number subsequent steps.
The issue of agents hallucinating standard React/Node commands (like
npx playwright test) rather than querying the actual repository environment causes frequent context corruption ("Agentic Amnesia"). Furthermore, agents fail to proactively scan.agent/skillsto discover expert workflows on boot.This ticket tracks the addition of a "Zero-Shot Discovery Mandate" to
AGENTS_STARTUP.md. SinceAGENTS_STARTUP.mdis invoked during the "Childhood" handshake of every new session, it is the perfect place to force an agent to runview_fileonpackage.jsonand.agent/skillsto ground itself before touching any code.Additionally, we will add an explicit instruction encouraging agents to recommend new skills if they identify recurring complex tasks that lack a documented workflow.
Task:
AGENTS_STARTUP.mdto include "Step 6: Discover the Repository Ecosystem & Skills".package.jsonfor available scripts..agent/skills/for capability workflows.