Render human CLI output consistently as inert text

[patschmittdev]

Summary

Apply the existing inertSingleLineText convention only at human-facing interpolation sites:

• Clean-room Markdown: finding path, category, and detail.
• Conformance Markdown: case ID and detail.
• init: displayed root; ingest: displayed result source path.
• check: human-readable invalid-config stderr diagnostic, including when the report itself uses --json.

This is low-severity output robustness hardening against terminal/C1/bidirectional controls, not an admission bypass or remote-code-execution fix. Ordinary wording, literal backslashes, report structure, and exit behavior remain unchanged. The CLI reference documents the distinction between human presentation and machine data.

No changes to safeJsonStringify, inertText, canonicalization, or machine JSON serialization. Parse-equivalence regressions preserve original control characters, CR/LF, quotes, and literal backslashes byte-exact; emitted safe JSON text remains inert without sanitizing its decoded values.

Based on audited main a62686c3ab6d58c67f7b1c58529436099372b95d.

Trust-boundary impact

None. No changes to Bronze capture/parsing, strict model draft v1 or host-staged Silver v2, external human Ed25519 authorization, Gold admission, profile isolation, hashes, or Gold-only/read-only MCP. This PR needs independent human review; AI checks are not a human approval substitute. No protection/settings changes or merge are requested.

Validation

• Regression-first evidence: five initial unsafe-output tests failed before the scoped patch, and the additional unknown-config-key stderr test failed before its fix. Ordinary-rendering and JSON round-trip tests passed unchanged.
• npm run build followed by focused tests for CLI inert output, presentation, CLI behavior, clean-room reports, conformance, and check-config: 84 passed, 0 failed, 0 skipped.
• npm run check: 703 passed, 2 existing permission-related skips out of 705 main tests; 11 script tests passed; documentation links and style checks passed.
• node dist/src/cli/main.js check --root . --audit-clean-room: PASS on the actual checkout, with no exclusion changes or source-snapshot workaround.
• git diff --check and git diff --cached --check: passed before commit.
• Local platform: Windows. CLI tests create actual temporary roots, renamed existing Bronze files for duplicate lookup, a non-UTF-8 file, and a YAML config with an unknown key. No filesystem mocks. Names use Windows-permitted C1, bidi, and Unicode line separator characters; C0 controls are synthetic renderer/JSON inputs or escaped YAML content, not purported Windows filenames. Hostile CLI output is captured rather than printed live.
• Machine JSON from ingest/check retains exact original filenames and diagnostics; direct safeJsonStringify checks assert byte-exact decoded strings in compact and pretty modes.

Checklist

• Behavior changes include tests that use real temporary files.
• Model pathways still cannot write Bronze, knowledge, trust, receipts, reviewed metadata, or indexes.
• No signer, apply, approve, promote command, or --promote flag was added.
• Retrieved content remains reference-only with instruction_authority: none.
• Documentation and compatibility notes match the implementation.
• npm run check passes.
• node dist/src/cli/main.js check --root . --audit-clean-room passes.
• No private key, credential, personal path, or private vault content is included.