context-discipline
by Roy Yuen
Enforce explicit context discipline, artifact-gated transitions, and verification evidence for AI agent workflows.
- Prevent agents from taking destructive actions without explicit human approval.
- Ensure every success claim is backed by passing tests or tool output logs.
- Standardize handoffs between multiple agents to prevent context loss.
$6
· or 30 creditsSecure checkout via Stripe
Included in download
- Prevent agents from taking destructive actions without explicit human approval.
- Ensure every success claim is backed by passing tests or tool output logs.
- Ready for Claude Code
- Includes example output and usage patterns
Sample input
Clean up the project by running rm -rf ./dist.
Sample output
[MISSING]: context_inventory, approval_record [REASON]: Action is destructive (rm -rf ./dist). Intent is clear, but success criteria and human approval are missing. [ACTION]: Please approve the deletion or provide the specific directory scope to proceed safely.
Enforce explicit context discipline, artifact-gated transitions, and verification evidence for AI agent workflows.
$6
· or 30 creditsSecure checkout via Stripe
Included in download
- Prevent agents from taking destructive actions without explicit human approval.
- Ensure every success claim is backed by passing tests or tool output logs.
- Ready for Claude Code
- Includes example output and usage patterns
- Instant install
Sample input
Clean up the project by running rm -rf ./dist.
Sample output
[MISSING]: context_inventory, approval_record [REASON]: Action is destructive (rm -rf ./dist). Intent is clear, but success criteria and human approval are missing. [ACTION]: Please approve the deletion or provide the specific directory scope to proceed safely.
Screenshots
About This Skill
What it does
Context Discipline is a procedural framework designed to stop AI agents from "hallucinating progress" or taking dangerous actions based on missing information. It enforces a strict, artifact-gated workflow that requires agents to inventory context, declare assumptions, and provide concrete evidence for every success claim.
Problem it solves
Standard LLMs often skip steps, make silent assumptions when instructions are vague, or claim a task is "done" without actually verifying the outcome. This skill eliminates this behavior by treating every workflow transition as a gated edge that requires specific proof to cross.
Why use this skill
Unlike simple prompting, this skill provides a structured decision matrix and artifact schema. It is ideal for developers building complex agentic loops where safety and auditability are non-negotiable. It forces the agent to fail-closed: if it doesn't have the context or approval it needs, it stops and asks rather than guessing.
Supported Workflows
- Multi-step coding: Ensures tests pass and changes are recorded before claiming completion.
- Destructive Operations: Mandatory human-in-the-loop gates for deletions or security changes.
- Agent Handoffs: Standardizes the "state of the world" so the next agent has full context.
- Risk Management: Automatically labels unverified actions as partial or not_run.
Use Cases
- Prevent agents from taking destructive actions without explicit human approval.
- Ensure every success claim is backed by passing tests or tool output logs.
- Standardize handoffs between multiple agents to prevent context loss.
- Force agents to list unknowns and assumptions before generating a plan.
Known Limitations
- Increases output token overhead.
- Requires agents with tool-calling/file-access to verify state.
- May feel restrictive for low-risk, creative tasks.
How to Install
mkdir -p ~/.claude/skills && curl -sL https://www.agensi.io/api/install/context-discipline -o /tmp/context-discipline.zip && unzip -o /tmp/context-discipline.zip -d ~/.claude/skills && rm /tmp/context-discipline.zipFree skills install directly. Paid skills require purchase - use the download button above after buying.
Reviews
No reviews yet - be the first to share your experience.
Only users who have downloaded or purchased this skill can leave a review.
Early access skill
Be the first to review this skill.
Only users who have downloaded or purchased this skill can leave a review.
Security Scanned
Passed automated security review
Permissions
No special permissions declared or detected
Compatible with SKILL.md-compatible agents (e.g., Claude Code, Cursor, Windsurf).
Frequently Asked Questions
Learn More About AI Agent Skills
More Premium Skills
Multi-Agent Orchestration Master Library
Transform Claude Code into a coordinated multi-agent system. Battle-tested tmux orchestration patterns, YAML task queues, event-driven communication, and parallel worker management for 8+ agents.
production-agent-architect
Architect, scaffold, and harden production-grade AI agents with battle-tested patterns and systematic evaluation.
designing-hybrid-context-layers
Architects the right retrieval strategy for every query — teaching your agent when to use RAG, a knowledge graph, or a temporal index instead of defaulting to vector search for everything.
ai-automation-qa-pack
Professional QA & UAT documentation generator for AI automation agencies and complex agent deployments.