v4.1.0: SKILL.md split (1490→250 lines), security fixes, divergence e…

…ngine

SKILL.md refactored: user-facing essentials only (250 lines). All architecture,
templates, composition rules, subteam deep-dive, output format specs, efficiency
rules moved to docs/ with progressive loading references.

Security: injection defense on Cross-Review, Devil's Advocate, Phase 5 templates.
Profile poisoning prevention. Provocateur injection defense. Scoped file access.

Divergence Engine: Provocateur, EXPLORE mode, preserve-if-unique triage,
creative disruption check, structural protections (adversarial immunity,
Socratic follow-ups, refutation resistance, Socratic Remainder, inverted
termination). GUIDE.md updated with EXPLORE mode docs.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

v4.0.0: Adaptive Intelligence — context-aware auto-configuration

Quorum now reads the project before configuring. No more flat-5 defaults.

Project Profiles: auto-generated on first run, persist context across runs.
Task Classification Gate: 4-dimension scoring (domain, certainty, scope,
artifact) → auto-selects mode, size, structure, rigor.
Config Transparency Block: shows reasoning before spawning agents.
Adaptive Output Templates: 5 formats matched to task type (audit, research,
dialectic, decision, org).

New flags: --yes, --quiet, --profile, --format.
Breaking: output format changes, default behavior changes.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

v3.2.0: Security hardening — remove Bash from manifest, add injection…

… defense

CRITICAL: Removed Bash/Write/Edit from manifest allowed-tools. These were
available to all spawned agents despite docs stating supervisor-only.

CRITICAL: Added XML tag injection defense with session boundaries and
agent-level prompt injection detection in all templates.

Also: credential detection patterns, path validation guidance, privacy
disclosure fixes, independence disclaimer for Phase 5, argument-hint
frontmatter, disable-model-invocation.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

v3.1.0: Research + Validation workflow — use Quorum as an epistemic q…

…uality gate

Add two-stage research pattern: gather with one swarm, validate with another.
Three copy-paste examples (two-stage, validate external research, resume session),
three-tier verdict output (VALIDATED/FLAGGED/BLOCKED), panel provenance, scope
disclaimers. Version bump 3.0.0 → 3.1.0.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>