v2.2.0 LTS - A Compete Workspace

Welcome to Soundingboard 2.2.0 (Long-Term Support)

Soundingboard 2.2.0 establishes a fully stable, robust architectural foundation for long-term production. This release bridges the gap between conversational prompt instructions and deterministic technical guarantees:

"The author is the novelist; the system bends to the author, never the author to the system."
"The human author always holds the red pen. The AI never unilaterally rewrites author prose."

In v2.2, this tenet moves from an advisory request into a mechanically enforced, unbypasable invariant.

---

What's New in v2.2.0

1. Mechanical Red Pen Enforcement Engine (scripts/redpen.js)

• The Bracket Invariant: The engine strips all [...] bracket tags from both old and new text. If the remaining author words are identical, the edit is recognized as an advisory Bracket Method diagnostic; if any author words are modified, added outside brackets, or deleted, the edit is intercepted.
• Mode Calibration (preferences.md):
  • Solo Mode: All edits to protected paths (manuscript/, writers_room/, stages/03_drafting/output/chapters/, preferences.md) prompt the author for confirmation (ask).
  • Hybrid Mode: Bracket-only diagnostic suggestions pass automatically (allow); word changes prompt for approval (ask).
  • Generative Mode: Bracket edits and new prose files pass automatically (allow); modifying existing author prose prompts for approval (ask).
• Frontmatter Discrimination: Cleanly separates YAML frontmatter metadata updates from guarded markdown body prose.
• Shell Loophole Defense: Blocks sneaky shell rewriting utilities (sed, echo >, mv, destructive redirection) and chained commands targeting protected paths.
• CLI Audit: node scripts/soundingboard.js redpen [--since <ref>] performs post-hoc Git diff audits to verify zero unbracketed prose changes.

2. Socratic Discovery Playbook #24 (/dig)

• Red Pen for Ideas: Helps authors discover unformed thoughts, thin motives, and thematic seams without the AI smuggling synthetic plot points or trope slop into the story.
• Core Contract:
  1. One question per turn.
  2. Zero smuggled ideas (no events, backstories, motives, or themes not stated by the author).
  3. The Seam Dig (probing gaps between atomics and author statements).
  4. The Meaning Ladder (Surface $\to$ Cause $\to$ Stakes $\to$ Meaning).
  5. Verbatim capture to writers_room/notes/dig-<target>-<date>.md.
• Deterministic Context Packer: node scripts/soundingboard.js pack dig [target] assembles target atomics, scene cards, and canon behind the scenes.

3. In-Repo Antigravity Integration (.agents/)

• Zero-Friction Setup: The entire studio configuration lives in the repository under .agents/. Authors simply open the project and trust the folder—no manual installation of global skills or rules required.
• In-Repo Rules: .agents/rules/agents_rule.md ensures AGENTS.md governs every conversation turn.
• Modern Skills as Slash Commands: All 19 skills (including /dig, /unstuck, /dialogue-heat, /sensory-bloom) are natively registered in .agents/skills/ as first-class slash commands and semantic agent tools.
• Lifecycle Hook Integration: .agents/hooks.json maps PreToolUse on write_to_file, replace_file_content, and run_command directly to scripts/redpen.js hook, and Stop to scripts/capture.js stop.
• Multi-Harness Skill Sync: node scripts/soundingboard.js sync-skills keeps .claude/skills and .agents/skills in perfect lockstep.

4. Literary Conversation Provenance Engine (scripts/capture.js)

• First-Class Provenance: Turns ephemeral agent chats into a tamper-evident, permanent part of the book's creation record.
• High-Contrast Visual Markdown:
  • Author: Flush-left ### ✍️ Author · HH:MM in high-contrast clean prose.
  • Agent: Indented blockquote > ### 🤖 Concierge · HH:MM.
  • Red Pen Audits: Docked badges > 🛡️ **Red Pen Event:** ....
• Publisher & Copyright Ready: Transcripts are automatically scrubbed of API keys, tokens, emails, and internal model thinking chains.
• Ledger Registry: Tracks every session in conversations/INDEX.md with SHA-256 integrity hashes to detect post-hoc tampering.

5. Harness Conformance Profile & 5-Minute Smoke Test

• Conformance Profile: docs/harnesses/harness-profile.md defines the 6 core capabilities required of any AI harness, eliminating fragile per-tool maintenance.
• Context Tax Guidelines: Explicit instructions for disabling idle MCPs (such as 15k-token browser servers) to preserve token budgets.
• Automated Smoke Test: tests/test_smoke.js verifies CLI execution, Red Pen gating, shell defenses, Socratic packing, and session capture in under 5 minutes.

---

Automated Verification & Stability

• Test Suite: 100% pass rate across 101 test scenarios (including 27 Red Pen invariant tests, 21 Conversation Provenance tests, 8 Smoke tests, and all 98 legacy regression tests).
• Dependencies: Zero external runtime dependencies (100% pure Node.js $\ge$ 18.0.0).