test(public-safety): centralize GitHub output sanitizer invariants

[JSONbored]

Parent roadmap: #82
Cross-link: Phase 6 safety work #269

Background

Public output safety is one of Gittensory's core trust boundaries. It should be tested centrally instead of relying on scattered checks.

Goal

Create a unified public/private safety proof suite for comments, checks, mention replies, PR packets, MCP public output, generated issue/PR text, extension public panels, and copied snippets.

Current behavior

Several sanitization tests exist, but the forbidden language boundary should be harder to regress as new surfaces are added.

Desired behavior

A single helper and fixture set validates every public-output producer against the same forbidden concepts. Private authenticated output remains separately tested so it can carry allowed private context without leaking it into public surfaces.

Implementation requirements

• Centralize forbidden concept checks.
• Add helper for public-output assertions.
• Cover GitHub comments, minimal checks, mention replies, PR packets, MCP public-safe text, extension public panels, generated issue/PR text, and copied public snippets.
• Keep private API/MCP tests separate.
• Avoid brittle exact-copy snapshots where semantic checks are better.

Public/private output boundaries

Public GitHub output must stay quiet-by-default and sanitized. It must not include private reviewability, scoreability, reward/risk, raw trust, wallet/hotkey, public score estimates, or farming language.

Acceptance criteria

• All public output producers use the shared test helper.
• Forbidden concept injection is caught.
• Private output can still include allowed private context in authenticated tests.
• New public surfaces must opt into the helper.
• Check-run annotations from feat(github-app): publish public-safe check-run annotations #272 are covered.
• Advanced Phase 6 outputs from feat(agent): add replayable decision snapshots #270, feat(agent): add repo-specific scenario simulator #271, feat(mcp): ship native resources prompts and output schemas #273, feat(settings): add repo policy compiler for focus manifests #277, and feat(analytics): add outcome learning loop for recommendation quality #278 have public/private boundary tests where applicable.

Testing requirements

• npm run test:ci must pass.
• Global coverage must remain at or above 97% for lines, statements, functions, and branches.
• Aim for 98%+ branch coverage locally to avoid CI variance.
• Add tests for every new branch, fallback path, sanitizer rule, and regression.
• Add invariant/property-style tests when behavior depends on sorting, gating, scoring, queue pressure, source-upload safety, public/private boundaries, or upstream drift.
• Public GitHub output must be tested against forbidden language: wallet, hotkey, raw trust score, payout, reward estimate, farming, private reviewability, private scoreability, and public score estimate.
• MCP/local tooling must prove source contents are not uploaded.

Additional test scenarios

• Comment fixture.
• Check-run fixture.
• Mention reply fixture.
• PR packet fixture.
• MCP public-safe fixture.
• Extension public-panel fixture.
• Generated issue/PR body fixture.
• Negative control for private API.