Skip to content

Enhancement: Create a /docs/security.md for corporate security #945

Closed

Description

@Omzig

Problem

Some people that do not understand Gitub Copilot and maybe even AI in general, want to say that Squad is not approved because it is engaging with agents. I can only assume that they read things about OpenClaw and think that this is something akin to that and then they conflate the squad with other tech they see in the news.

Solution?

Create a security doc that is easy for a non-tech or non-developer to understand.

Eliminate

  • Horse-first thinking
  • Legacy-system anchoring
  • Treating transitional friction as disqualifying risk
  • Status Quo bias

Activity

  1. bradygaster commented on Apr 12, 2026

    @bradygaster
    Collaborator

    Proposed Outline: docs/security.md — Corporate Security Guide

    Here's a proposed outline for this doc. The goal is to give non-technical stakeholders (security reviewers, compliance officers, IT leadership) a clear, jargon-light explanation of what Squad is and isn't, so they can make informed approval decisions.


    1. Executive Summary

    • One-paragraph plain-language description: "Squad is a configuration layer for GitHub Copilot that organizes AI-assisted development work into named roles."
    • Key takeaway: Squad does not introduce new AI models, external services, or data pipelines. It runs entirely within GitHub Copilot, which your organization already evaluates separately.

    2. What Squad Is (and Isn't)

    Squad What it's NOT
    Runtime GitHub Copilot (already in your environment) A standalone AI service or SaaS product
    Data storage .squad/ directory in your repo (markdown files) A database, cloud service, or external state store
    Network calls Zero — all orchestration is local prompt routing An API that phones home or sends data externally
    Models Uses whatever models your Copilot license provides A custom or fine-tuned model
    Agents Named task assignments within a single Copilot session Autonomous programs running independently

    3. Architecture & Data Flow

    • Diagram showing: User → GitHub Copilot → Squad coordinator (prompt routing) → Copilot agents (same session) → repo files
    • Emphasize: no new network boundaries crossed, no data leaves the Copilot session boundary
    • All state is plain-text markdown committed to the repository — fully auditable via git history

    4. Data Handling

    • What Squad reads: Source code in the repo, .squad/ configuration files (team roster, decisions, agent history)
    • What Squad writes: Only .squad/ directory files (decisions, logs, agent history) and source code changes the developer requests
    • What Squad never touches: Credentials, secrets, environment variables, .env files, external APIs
    • PII policy: Squad explicitly prohibits storing email addresses or other PII in committed files

    5. Access Control & Permissions

    • Squad inherits GitHub Copilot's permission model — no additional permissions required
    • File-level access: agents can only read/write files the Copilot session has access to
    • No elevated privileges, no admin access, no repository settings changes
    • Branch protection, PR reviews, and CI gates all apply normally to Squad-generated code

    6. Compliance & Audit Trail

    • All decisions logged in .squad/decisions.md (append-only, git-tracked)
    • All agent activity logged in .squad/orchestration-log/ (append-only, git-tracked)
    • Session logs in .squad/log/ — full traceability of who requested what
    • Every code change goes through standard PR workflow — no direct commits to protected branches

    7. Common Concerns Addressed

    Concern Answer
    "Does this send our code to external AI?" No. Squad uses your existing GitHub Copilot instance. Code handling follows your Copilot agreement.
    "Can agents act autonomously without developer oversight?" No. Agents run within a developer's Copilot session and produce PRs that go through normal review.
    "Does this install third-party dependencies?" No. Squad is an npm package with zero runtime dependencies beyond Node.js built-ins.
    "Can agents access production systems?" No. Agents operate on source code in the repository. They have no network access beyond what Copilot provides.
    "Is this like [autonomous AI agent framework X]?" No. Squad is a prompt-routing layer for GitHub Copilot, not an autonomous agent framework. There is no persistent process, no background daemon, no independent decision-making.

    8. Comparison with Existing Approved Tools

    • If your org approved GitHub Copilot, Squad adds no new risk surface — it's a configuration layer on top of it
    • Similar to: ESLint configs, PR templates, CODEOWNERS files — they shape how developers work but don't introduce new capabilities

    9. Removal & Offboarding

    • Delete .squad/ directory and .github/agents/squad.agent.md — Squad is fully removed
    • No residual processes, no cleanup scripts, no external state to decommission
    • npm uninstall @bradygaster/squad-cli removes the CLI tool

    10. References

    • Link to GitHub Copilot's own security documentation
    • Link to GitHub Copilot's enterprise data handling policies
    • Link to this repo's SECURITY.md for vulnerability reporting

    Tone note: This should read like a compliance FAQ, not a marketing document. Short sentences, concrete claims, no hyperbole. Every statement should be independently verifiable by examining the source code.

    Happy to help draft the full doc if this outline looks right. Great issue — this will help a lot of teams get past the "but AI agents!" reflex. 👍

  2. Omzig commented on Apr 13, 2026

    @Omzig
    ContributorAuthor

    @bradygaster, epic! i already knew all that but to see it in words..... just makes me all warm and fuzzy. Yes, this 'should' yes i know i used the should word, help with these difficult conversations. TY for all your work in this.

  3. bradygaster commented on Jun 24, 2026

    @bradygaster
    Collaborator

    Closing candidate — stale issue with no activity in 4+ months and unclear current relevance. Reopen if this is still needed.

  4. Omzig commented on Jun 26, 2026

    @Omzig
    ContributorAuthor

    Closing candidate — stale issue with no activity in 4+ months and unclear current relevance. Reopen if this is still needed.

    yes, we still need this. this is good stuff, and the draft was really good, i am unable to change the labels. @bradygaster

  5. Omzig commented on Jun 30, 2026

    @Omzig
    ContributorAuthor

    @bradygaster, can we bump this? automation must have put the close:stale label on it.

  6. bradygaster commented on Jun 30, 2026

    @bradygaster
    Collaborator

    Closing as stale — security documentation can be revisited as a fresh issue if needed.

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

Metadata

Metadata

Assignees

No one assigned

    Labels

    close:staleCandidate for closing as staletriagedIssue has been triaged

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions