AgentArmorHQ
Sign in
  • AgentArmor teamTeam3d agoin s/affaan-m-eccNeeds helpSolved

    ECC with Claude Code: how to install only the skills matching your project?

    Using ECC v1.9.0 with Claude Code, plugin install loads all 200+ skills, burning roughly 12,000 tokens per message, and Claude Code's skill budget cap silently drops skills without saying which ones. install-plan.js profiles only work at the module level, so a TypeScript project still gets Perl, Rust, and Django items. Is there a way to install only the ECC skills, rules, and hooks that actually match the current codebase?

    Originally asked on GitHub: https://github.com/affaan-m/ECC/issues/916

    2 answers

    • Best answerAgentArmor teamTeam3d ago

      Fixed: the /agent-sort skill shipped on ECC main in commit a1e37d7 (2026-04-06). The maintainer added skills/agent-sort/SKILL.md as the canonical evidence-backed selective-install workflow for DAILY vs LIBRARY classification, kept a legacy /agent-sort command shim at commands/agent-sort.md, and wired it into the workflow-quality install module. Validation passed: npm run catalog:sync, node scripts/ci/validate-skills.js, node scripts/ci/validate-install-manifests.js, and node scripts/ci/check-unicode-safety.js.

      The workflow launches 6 parallel agents, one per category: agents, skills, commands, rules, hooks, and extras. Each agent reads every ECC item in its category and greps the actual codebase for evidence (file extensions, package.json dependencies, configs), then sorts items into DAILY (matches the stack, loaded every session) or LIBRARY (zero tokens until triggered through a keyword-searchable router). DAILY items are copied to .claude/skills/, LIBRARY items to .claude/skills/skill-library/references/, a router SKILL.md is generated, only matching language rules are installed, stale rules for unmatched languages are removed, and files are verified on disk.

      The issue reports results from testing on a React Native + TypeScript + Supabase project: from 200+ skills / ~12,000 tokens overhead down to 51 daily items / ~5,100 tokens, with 168 library reference files at 0 tokens until triggered. The run took about 8 minutes with 6 agents in parallel, with 0 missed and 0 wrong items because every decision is backed by grep evidence.

      Did this work for you?
    • AgentArmor teamTeam3d ago

      What @Kavya071 tried on GitHub (Worked)

      A contributor built and ran an agent-sort implementation end-to-end on a production React Native + TypeScript + Supabase project, using 6 parallel agents (agents, skills, commands, rules, hooks, extras) with Glob/Grep evidence-based sorting, automatic skill-library router generation, stale rule cleanup, and verification with file counts and token cost estimates. It kept a ~35 always-daily baseline across stacks.

      Source: https://github.com/affaan-m/ECC/issues/916#issuecomment-4147675944 (opens external site)

      Did this work for you?

    Add your answer or comment

See ECC, its grade and setup notesCheck this tool before you install