entroly-context-control

Surgically select, compress, and recover codebase context using Entroly's MCP tools. Slashes token usage by up to 95% with Merkle-exact recovery receipts and WITNESS verification.

SKILL.md

GitHub

Entroly Context Intelligence Skill

Use this skill to optimize context, query the codebase with token budgets, compress bulky tool outputs, and recover omitted code byte-for-byte using the Entroly MCP server.

Core Workflows

1. Token-Budgeted Codebase Selection (optimize_context)

When answering a complex question or preparing a refactor across multiple files:

  • Call optimize_context with your natural language query and a token budget (e.g., budget=4000 or budget=8000).
  • Entroly executes a knapsack optimization algorithm that ranks AST fragments by relevance, entropy, and dependency relations.
  • Returns surgical code fragments, line numbers, and a cryptographic Merkle receipt.

2. Semantic Code Recall (recall)

When searching for specific symbols, implementations, or interfaces:

  • Call recall(query="...") with top_k=5 or top_k=10.
  • Returns ranked pointers with filenames, line numbers, and snippet digests.

3. Reversible Output Compression (compress & recover)

When dealing with large terminal dumps (build traces, test logs):

  • Call compress(data=..., budget=...) to distill down critical errors and signals while stripping noise.
  • Every compression produces a content-addressed SHA-256 handle.
  • To restore the original uncompressed text at any time, call recover(handle="...").

4. Codebase Health & Architecture Diagnostics (health)

To assess codebase quality before or after major changes:

  • Call health() to inspect clone pairs, dead symbols, god files, and architecture violations.

5. Grounding & Hallucination Verification (witness_verify)

To verify an agent claim or proposed code against source evidence:

  • Call witness_verify(claim=..., evidence=...) for local NLI-backed verification.