absorb-local-learnings

Internal repository workflow to inspect machine-global rules and agent learnings (~/.gemini, ~/.claude) and safely absorb portable engineering invariants into repository rules, skills, or agents while

SKILL.md

GitHub

Skill: Absorb Local Learnings (Internal Maintenance Guide)

This internal repository skill governs the extraction, classification, sanitization, and absorption of operational heuristics, AI runtime lessons, and coding guidelines learned across local environments (~/.gemini/config/, ~/.claude/) into the versioned amiga-ia repository.


1. Trigger Conditions

Run this workflow when:

  • The user asks to reconcile or sync local AI rules/learnings with the repository.
  • A debugging session in a sibling repository produced a new generalizable engineering rule or anti-hallucination guardrail stored in ~/.gemini/config/rules/ or global AGENTS.md / CLAUDE.md.
  • Preparing a minor or major release where local operational invariants should be codified for the entire distributed community.

2. Phase 0: Discovery & Diff Inventory

Inspect all global local configuration directories:

  1. Google Antigravity:
    • ~/.gemini/config/rules/*.md
    • ~/.gemini/config/AGENTS.md
  2. Anthropic Claude Code:
    • ~/.claude/CLAUDE.md
    • ~/.claude/rules/*.md (if present)
  3. OpenAI Codex:
    • ~/.codex/ and ~/.agents/skills/

Compare file timestamps and structural contents against existing repository assets in rules/, skills/, agents/, and repo-level AGENTS.md.


3. Phase 1: Privacy, Employer & Personal Isolation Gate (STRICT FILTER)

Every discovered local rule, instruction, or guideline must pass through the Sanitization and Portability Filter before absorption.

Forbidden Categories (NEVER ABSORB / LEAVE STRICTLY LOCAL)

  1. Local Machine & Filesystem Specifics:
    • Machine-specific absolute paths (e.g., user profile paths, home directories, local application data directories), local user-level binary installations, or environment-specific toolchain path references.
  2. Workplace & Employer Specifics:
    • Corporate repositories, proprietary client workflows, employer-specific accounts/tokens, internal company databases, or private organizational context files.
  3. Personal Aesthetics & Bespoke Design Systems:
    • Personal brand design systems, signature portfolio aesthetics, author-specific color palettes, custom line art or watermark directives, or personal homepage references.

Approved Portable Value (ELIGIBLE FOR ABSORPTION)

  1. Autonomous AI Orchestration & Guardrails:
    • Subagent anti-hallucination protocols, mandatory live verification gates (releases, git pushes, assets).
    • Execution patterns addressing platform sandboxes (e.g., Antigravity Pattern A vs. Pattern B).
    • Anti-bottleneck parallelization rules and investigation bifurcation.
  2. Cross-Platform Engineering Standards:
    • Compilation and version bump ordering in release pipelines.
    • Native toolchain cache pruning (cargo clean, dotnet clean, ./gradlew clean) and artifact hygiene.
    • Desktop application stability invariants (WinUI 3 / WPF / MAUI exception handling, XAML security, UI thread dispatching).
  3. Universal Documentation & Testing Hygiene:
    • Zero country flags policy.
    • Zero static test counts (atemporal quality gates).
    • Path privacy rules and clean badge specifications.

4. Phase 2: Destination Triage Matrix

Classify each approved item into the appropriate repository layer:

Target Destination Purpose Standards & Conventions
rules/<name>.md Global declarative rules and operational guardrails for Antigravity & Claude. Written in clear domain language (English for technical rules, or Spanish for shared documentation standards). Must include frontmatter trigger: always_on where applicable.
skills/ami-<name>/ Reusable, multi-step interactive workflows or developer procedures. Must follow ami-<verb>-<object> action naming, YAML frontmatter with canonical allowed-tools, under 250 lines, procedural clarity, no emojis.
agents/ami-<name>.md Autonomous persona definitions requiring permission isolation or context offloading. Must follow ami-<role> naming, describe responsibilities and tool authorizations without duplicating skill step-by-step logic.
AGENTS.md (Repo Root) Internal development rules governing the amiga-ia codebase itself. Numbered rules (e.g., Rule 24: Anti-Bottleneck). English only.

5. Phase 3: Interactive Confirmation Gate

CONFIRMATION 1 - Triage Review: Before creating or modifying any file, construct and present a classified summary table to the user:

  • Files to Absorb: Target path, source path, rationale, and target architectural layer.
  • Files Excluded: Specific reason for exclusion (e.g., private paths, employer context, personal design system).

STOP and await explicit user approval. Never apply changes speculatively.


6. Phase 4: Execution & Verification Gate

Once approved:

  1. Apply modifications using replace_file_content or write_to_file.
  2. Ensure all newly added public skills or agents adhere strictly to AMI conventions:
    • Zero emojis in prompts and instructions.
    • English source code, docstrings, and commits.
    • Symmetric bilingual documentation where required.
  3. Execute the repository verification gates:
    node bin/setup.js --doctor
    npm test
    
  4. Verify that git status reflects clean, isolated modifications strictly conforming to repository architectural layering.