Skip to content

feat: add agent-safety instructions and governance reviewer agent#757

Merged
aaronpowell merged 2 commits into
github:mainfrom
imran-siddique:add-agent-governance-instructions
Feb 18, 2026
Merged

feat: add agent-safety instructions and governance reviewer agent#757
aaronpowell merged 2 commits into
github:mainfrom
imran-siddique:add-agent-governance-instructions

Conversation

@imran-siddique
Copy link
Copy Markdown
Contributor

Summary

Adds two complementary governance contributions:

1. instructions/agent-safety.instructions.md

Custom instructions that guide Copilot toward safe AI agent development patterns:

  • Tool access controls: Explicit allowlists, blocklist dangerous operations, human-in-the-loop for high-impact tools
  • Content safety: Pre-flight intent scanning, argument filtering for credentials/PII
  • Multi-agent safety: Trust boundaries, most-restrictive policy propagation
  • Audit & observability: Append-only logging, JSON Lines export
  • Framework-specific notes: PydanticAI, CrewAI, OpenAI Agents, LangChain, AutoGen
  • Common mistakes: Anti-patterns to avoid (output-only guardrails, self-modifying policies, stale trust)

2. agents/agent-governance-reviewer.agent.md

A specialized Copilot agent that reviews code for governance gaps:

  • Checks for missing governance decorators on tool functions
  • Verifies input scanning for threat signals
  • Looks for hardcoded credentials in agent configurations
  • Confirms audit logging exists
  • Reviews trust boundaries in multi-agent systems

Context

These pair with the agent-governance skill (PR #755) and governance-audit hook (PR #756) to provide a complete governance toolkit for Copilot users building AI agent systems.

- instructions/agent-safety.instructions.md: Guidelines for building safe,
  governed AI agent systems (tool access controls, content safety,
  multi-agent safety, audit patterns, framework-specific notes)
- agents/agent-governance-reviewer.agent.md: Expert agent that reviews code
  for governance gaps and helps implement policy enforcement

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Copilot AI review requested due to automatic review settings February 18, 2026 21:30
Copy link
Copy Markdown
Contributor

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Adds new governance-focused guidance assets (one instruction set + one review agent) to help Copilot users build safer, more auditable AI agent systems, and indexes them in the docs.

Changes:

  • Added Agent Safety & Governance custom instructions for safe tool-calling and multi-agent patterns.
  • Added Agent Governance Reviewer custom agent to review codebases for governance gaps.
  • Updated instruction/agent documentation indexes to include the new resources.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

File Description
instructions/agent-safety.instructions.md New safety/governance instruction set for agent/tool-calling implementations.
agents/agent-governance-reviewer.agent.md New agent definition focused on governance and auditability reviews.
docs/README.instructions.md Adds the new instruction to the published instructions index.
docs/README.agents.md Adds the new agent to the published agents index.

Comment thread instructions/agent-safety.instructions.md
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
@aaronpowell aaronpowell merged commit f880e12 into github:main Feb 18, 2026
3 checks passed
@imran-siddique
Copy link
Copy Markdown
Contributor Author

Review comment addressed in commit 03290d7:

applyTo field: Added applyTo: '**' to agent-safety.instructions.md frontmatter to scope the instruction globally, matching repo conventions.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants