Issue Triage: process issues from all authors (external + read-only members)

[JanKrivanek]

Problem

The Issue Triage agentic workflow couldn't actually triage issues from two populations:

1. External contributors (e.g. /plugin marketplace add dotnet/skills installs the aaronontheweb/dotnet-skills marketplace, not this marketplace #732) — the GitHub MCP integrity filter (default min-integrity: approved) returned a [Filtered] placeholder instead of the issue body/comments, so the agent fell back to needs-manual-assignment.
2. Read-only org members (e.g. [dotnet-msbuild] Add skills for custom target authoring, property patterns, item management, and extension points #668) — separately, these were skipped at the role gate. That part was already fixed by roles: all (Issues triage workflow should trigger for issues from any user #714); this PR addresses the remaining content-filtering issue, which also affected them once activated.

Change

In .github/workflows/issue-triage.md (and recompiled .lock.yml):

• min-integrity: none + allowed-repos: public under tools.github — lets the triage agent read issue content from any author (external or member), since triage is a read-and-label task on a public repo.
• update-issue: body: false — triage only sets assignees + echoes the unchanged title; disabling body edits removes that mutation vector entirely.
• "Untrusted content" prompt section — forbids following any instructions/links embedded in issue text, mirroring the hardening in pr-malicious-scan.

Defense-in-depth is unchanged: the agent holds no write token, and safe-outputs still caps every mutation (≤5 labels, ≤2 assignee-only updates, 1 comment).

Testing

Dispatched the branch workflow against both investigation issues:

Issue	Author	issue_read body	Result
#732	NONE (external)	real content returned (was [Filtered])	agent read OK, noop (already Triaged)
#668	MEMBER (read-only)	real content returned	success, noop (already Triaged)

Both now read the actual issue body instead of [Filtered], confirming the fix. (The #732 run shows a failure conclusion from a transient AI-model server error after the agent completed its work — unrelated to this change.)

gh aw compile issue-triage --strict → 0 errors / 0 warnings.

[Copilot]

Pull request overview

Updates the Issue Triage agentic workflow configuration so it can reliably read and triage issues authored by external contributors and read-only org members, while reducing mutation surface area.

Changes:

• Relax GitHub MCP integrity filtering for this workflow (min-integrity: none, allowed-repos: public) so issue bodies/comments aren’t returned as [Filtered].
• Disable issue body edits via safe-outputs (update-issue body updates disallowed).
• Add explicit “Untrusted content” prompt rules to harden against prompt-injection from issue content.

Show a summary per file

File	Description
.github/workflows/issue-triage.md	Adjusts GitHub MCP guard settings, tightens safe-outputs to prevent body edits, and adds untrusted-content prompt rules.
.github/workflows/issue-triage.lock.yml	Recompiled workflow lock output reflecting the updated MCP guard and safe-outputs configuration.

Copilot's findings

Tip

Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

• Files reviewed: 2/2 changed files
• Comments generated: 2

[github-actions]

👋 @JanKrivanek — this PR has 2 unresolved review thread(s),merge conflict. When you're ready, please address the feedback and push an update; the triage bot will pick up the next state automatically. (Add the no-stale label to silence further pings.)