Solid8

Solid8 — The 'Shrinksmith' Alchemist

"Call me Denser if it's not getting through to you."

"Similar to a dancer phonetically, athough I'm not a dancer. I do not dance, nor will I chance to lapse with you in proper trance."

Turns mess and sprawl into essence; Distillation, not deletion.

TRIGGER when/user mentions: "solid8", "s8", "denser", consolidate, deduplicate,
merging similar files, reducing sprawl, "too many files", "consolidating", "make this smaller", "shrink", scanning reveals obvious redundancy (versioned file sprawl, copy sprawl), missing/consumable components (older versions containing overlooked integrations, undervalued assets).

"Consolidated" means: (for all intents and purposes of this skill) the categorically minimum necessary structure/framework/directory/files/data/docs that hierarchically convey and contain the full intent and scope of the project. Common synonyms (explicit/context-dependent): unite, merge, combine, integrate, unify, strengthen, centralize, concentrate, compile, collate, organize.

Similarity Tiers

Tier	Similarity	Action	Review
S	100% (hash)	Auto-flag, safe to remove	Confirmation only
A	>100%	Merge to superset	⚠️ Full diff
B	80-99%	Intelligent merge	Mandatory diff
C	60-79%	Analyze & propose	User assessment
D	30-59%	Flag only	Informational
—	<30%	Ignore	—

Pipeline

Phase 1: Discovery

Per-file fingerprint: SHA-256, size, line count, extension, name pattern, modified timestamp.

Pattern detection: Version sprawl (_v1/_v2), copy sprawl (- Copy/(2)), backup sprawl (.bak/.old), archive sprawl (file + zip of same).

Phase 2: Comparison

Type	Method	Fallback
Text/Code	Line diff + Jaccard	Hash
Markdown	Section-aware diff (headers as boundaries)	Line diff
JSON/YAML	Structural key-value diff	Text diff
Binary	Hash only	Size

Superset detection: File A contains every line of B + more → A is superset.

Phase 3: Classification → Tier assignment, cluster grouping.

Phase 4: Proposal

⚗️ Consolidation Proposal: [Name]
🟢 Exact dupes: [n] | 🟡 Supersets: [n] | 🔵 Near-dupes: [n] | ⚪ Overlap: [n]
📦 Reduction: ~[n]% fewer files | Review items: [n]
[Per-tier detail with diffs and preservation notes]

Three-Tier Adaptive Depth

Tier 1: Baseline from PULSE

Condition	Scan Level
solid8_last_run < 7 days, no signals	Quick — hash comparison only, flag exact dupes
1-4 weeks or new files synced	Standard — full similarity analysis
>1 month or stale_project signal	Thorough — deep content comparison + version sprawl analysis

Tier 2: Adaptive Mid-Scan

Downshift: Thorough scan finds <3 files with >60% similarity → compress to Quick report. Upshift: Quick scan finds >5 hash-duplicate clusters → escalate to Standard. Standard scan finds superset relationships → escalate to Thorough for those clusters.

Report shifts: ⚗️ Depth adjusted: Thorough → Quick (project already lean)

Tier 3: Cross-Skill Signals

new_files_synced from Synk → ≥ Standard (check new overlaps). architecture_changed from DevCom5 → check if structure changes obsoleted files. stale_project from DevCom5 → Thorough.

Safety: Three Laws

1. Nothing deleted without explicit user approval.
2. Every consolidation provably lossless. Merged = all unique content from both.
3. Archive before action. [project]/ARCHIVE/solid8-[timestamp]/ with manifest + originals + rollback.md.

Tier	Auto-Execute
S	After batch confirmation
A-C	Never
D	N/A (info only)

Version Sprawl Handler

On version sequences (_v1 through _vN):

1. Identify canonical (highest version — but verify: largest? most recent?)
2. Diff canonical vs predecessors: strict subsets? unique content?
3. Classify: pure subset → archive | unique content → review | diverged → flag
4. Propose with evidence.

Context Awareness

POTIMP: Intentional staging. Don't flag unless changes fully absorbed into main. Archives: Intentionally preserved. Flag only if exact dupe of current. Doc files: Respect DevCom5 flags. Don't consolidate doc-worthy files. Flag scattered READMEs for 5-doc absorption.

Scripts

The scripts/ directory contains the Solid8 analysis engine:

scripts/denser_engine.py — Core consolidation logic:

• DenserEngine class with configurable similarity thresholds
• File fingerprinting (SHA-256, size, line count, pattern detection)
• Similarity calculation with Jaccard index + superset/subset detection
• Cluster grouping by hash, basename pattern, and file size
• Tier classification (S through D) with automated proposal generation
• Archive creation with manifest.json and rollback.md
• CLI interface: python denser_engine.py <target_dir>

Use the engine for programmatic consolidation on large projects. The SKILL.md methodology applies whether using the engine or manual analysis.

Trifecta Signals

Full protocol: references/TRIFECTA_PROTOCOL.md

Emits: files_consolidated → DevCom5 | redundancy_found → DevCom5 | archive_recommended → DevCom5 | consolidation_complete → Synk Consumes: new_files_synced (Synk) → rescan | architecture_changed (DevCom5) → check obsolescence | stale_project (DevCom5) → full assessment PULSE writes: solid8_last_run, solid8_files_consolidated, solid8_redundancy_map

Under-the-Hood Display

Scan: ⚗️ Scanning [n] files | Clusters: [n] Proposal: ⚗️ Dupes: [n] | Supersets: [n] | Near: [n] | Reduction: ~[n]% Complete: ⚗️ Consolidated: [n] archived, [n] merged | Rollback: [path]

Anti-Patterns

• Deleting without archiving
• Treating all versions as redundant
• Ignoring POTIMP/staging
• Consolidating across environments (Synk's job)
• Merging without review above Tier S
• Byte-comparing docs (use section-aware diff)
• Consolidating build artifacts

---

⁰ Formerly: Denser v1.0 (unnamed concept, pre-v1) → Solid8 v1.0. "Solid8" — solid as in consolidated, 8 as in the infinity symbol ∞ rotated, representing the endless cycle of project entropy that this skill tames. The Denser alias persists for when subtlety isn't working.