Skip to content

Skills Overview

The Mistaber encoding plugin provides six specialized skills that guide the AI through the complete halachic encoding pipeline. Each skill handles a specific phase of the encoding process, with mandatory human checkpoints ensuring quality and accuracy.

Skill Architecture

graph TB
    subgraph "Encoding Pipeline Skills"
        direction LR
        CP[corpus-prep] -->|Checkpoint 1| HE[hll-encode]
        HE -->|Checkpoint 2| VA[validate]
        VA -->|Checkpoint 3| RE[review]
        RE -->|Checkpoint 4| CO[commit]
    end

    subgraph "Documentation Skill"
        WG[workflow-guide]
    end

    WG -.->|Documents| CP
    WG -.->|Documents| HE
    WG -.->|Documents| VA
    WG -.->|Documents| RE
    WG -.->|Documents| CO

The Six Skills

1. Corpus Preparation (corpus-prep)

Purpose: Fetch and organize source texts from Sefaria, build complete derivation chains, identify machloket.

Checkpoint: Source Review (Checkpoint 1)

Aspect Details
Phase 1 of 5
Trigger Phrases "prepare corpus", "fetch sources", "start encoding", "begin encoding"
Prerequisites Sefaria MCP available, valid seif reference
Outputs corpus-report.md, corpus-sources.yaml, corpus-chain.mermaid, corpus-questions.yaml
Requires Approval Yes - before proceeding to encoding

Full Documentation


2. HLL Encoding (hll-encode)

Purpose: Transform approved corpus into HLL/ASP rules with proper world scoping and source attribution.

Checkpoint: Rule Review (Checkpoint 2)

Aspect Details
Phase 2 of 5
Trigger Phrases "encode rules", "write HLL", "create ASP rules", "translate to HLL"
Prerequisites Checkpoint 1 approved, corpus artifacts available
Outputs base.lp (shared), world-specific .lp files, encoding-report.md, encoding-mapping.yaml
Requires Approval Yes - before proceeding to validation

Full Documentation


3. Validation (validate)

Purpose: Compile rules, run semantic checks, execute test scenarios, verify machloket encoding.

Checkpoint: Test Review (Checkpoint 3)

Aspect Details
Phase 3 of 5
Trigger Phrases "validate encoding", "run tests", "compile rules", "check rules"
Prerequisites Checkpoint 2 approved, encoded .lp files exist
Outputs validation-report.md, validation-results.yaml, test-scenarios.yaml
Requires Approval Yes - before proceeding to review

Full Documentation


4. Review (review)

Purpose: Assemble comprehensive review package for final human approval.

Checkpoint: Final Approval (Checkpoint 4)

Aspect Details
Phase 4 of 5
Trigger Phrases "review encoding", "prepare review", "final review", "create review package"
Prerequisites Checkpoint 3 approved, all tests passing
Outputs review-package.md (comprehensive)
Requires Approval Yes - before proceeding to commit

Full Documentation


5. Commit (commit)

Purpose: Finalize and commit approved encoding to the repository.

Checkpoint: None (Terminal Skill)

Aspect Details
Phase 5 of 5
Trigger Phrases "commit encoding", "finalize encoding", "save encoding", "complete encoding"
Prerequisites All checkpoints (1-4) approved
Outputs Git commit, updated manifest, archived session
Requires Approval N/A - all approvals already obtained

Full Documentation


6. Workflow Guide (workflow-guide)

Purpose: Provide comprehensive documentation of the entire encoding pipeline.

Checkpoint: None (Documentation Only)

Aspect Details
Phase N/A
Trigger Phrases "how does encoding work", "what is the workflow", "explain pipeline"
Prerequisites None
Outputs None (informational only)
Requires Approval N/A

Full Documentation

Skill Invocation

Invocation Patterns

Skills are invoked automatically when the AI recognizes trigger phrases in user messages:

User: "Prepare corpus for YD 87:3"
→ corpus-prep skill invoked

User: "Encode the rules"
→ hll-encode skill invoked

User: "Run validation"
→ validate skill invoked

User: "Prepare the review package"
→ review skill invoked

User: "Commit the encoding"
→ commit skill invoked

Explicit Invocation

Skills can be explicitly invoked using the plugin prefix:

User: "Use mistaber:corpus-prep for YD 87:3"
User: "Invoke mistaber:validate"

Invocation via Orchestrator

The encoding-orchestrator agent coordinates all skills automatically:

User: "Encode YD 87:3"
→ encoding-orchestrator activates
→ Guides through corpus-prep → hll-encode → validate → review → commit
→ Pauses at each checkpoint for human approval

Skill Dependencies

Linear Pipeline

Skills form a strict linear pipeline with checkpoint enforcement:

graph LR
    A[Start] --> B{corpus-prep}
    B -->|Checkpoint 1| C{hll-encode}
    C -->|Checkpoint 2| D{validate}
    D -->|Checkpoint 3| E{review}
    E -->|Checkpoint 4| F{commit}
    F --> G[Complete]

    style B fill:#90EE90
    style C fill:#87CEEB
    style D fill:#DDA0DD
    style E fill:#F0E68C
    style F fill:#98FB98

Checkpoint Requirements

Transition Requires
Start → corpus-prep None
corpus-prep → hll-encode Checkpoint 1 approved
hll-encode → validate Checkpoint 2 approved
validate → review Checkpoint 3 approved
review → commit Checkpoint 4 approved

Hook Enforcement

The phase-gate hook enforces dependencies:

# Attempting hll-encode without corpus-prep approval
# → BLOCKED: Checkpoint Not Approved

# Attempting commit without review approval
# → BLOCKED: Review Not Approved

Skill Artifacts

Artifacts by Phase

Phase Working Artifacts Final Output
corpus-prep corpus-report.md, corpus-sources.yaml, corpus-chain.mermaid, corpus-questions.yaml -
hll-encode encoding-report.md, encoding-mapping.yaml corpus/yd_{siman}/base.lp, worlds/*.lp
validate validation-report.md, validation-results.yaml, test-scenarios.yaml -
review review-package.md -
commit - Git commit, manifest update

Artifact Locations

.mistaber-artifacts/                    # Working artifacts
├── corpus-report-YD-{siman}-{seif}.md
├── corpus-sources-YD-{siman}-{seif}.yaml
├── corpus-chain-YD-{siman}-{seif}.mermaid
├── corpus-questions-YD-{siman}-{seif}.yaml
├── encoding-report-YD-{siman}-{seif}.md
├── encoding-mapping-YD-{siman}-{seif}.yaml
├── validation-report-YD-{siman}-{seif}.md
├── validation-results-YD-{siman}-{seif}.yaml
├── test-scenarios-YD-{siman}-{seif}.yaml
├── review-package-YD-{siman}-{seif}.md
└── source-chain-log.yaml

mistaber/ontology/                      # Final output
├── corpus/yd_{siman}/
│   ├── base.lp                         # Shared definitions
│   └── manifest.yaml                   # Siman metadata
├── worlds/
│   ├── mechaber.lp                     # Updated with new rules
│   └── rema.lp                         # Updated with new rules
└── tests/yd_{siman}/
    ├── seif_{seif}_test.yaml
    └── seif_{seif}_metadata.yaml

docs/encoding-sessions/                 # Archive
└── yd_{siman}_{seif}_{date}/
    └── [all working artifacts]

Session State Integration

State Tracking

Each skill updates .mistaber-session.yaml:

current_phase: hll-encode
target_seif: "YD:87:3"
started: 2026-01-25T10:00:00Z
checkpoints:
  corpus-prep:
    status: approved
    approved_by: human
    timestamp: 2026-01-25T10:30:00Z
    artifacts:
      - .mistaber-artifacts/corpus-report-YD-87-3.md
    complexity_score: 6

  hll-encode:
    status: pending_review
    artifacts:
      - mistaber/ontology/corpus/yd_87/base.lp

  validate:
    status: not_started

  review:
    status: not_started

Status Values

Status Meaning
not_started Phase not yet begun
in_progress Phase actively executing
pending_review Awaiting human approval
approved Human approved, can proceed
tests_passed (validate only) Tests completed successfully
tests_failed (validate only) Tests have failures

Skill Communication

Inter-Skill Data Flow

Skills communicate through artifacts:

graph TD
    CP[corpus-prep] -->|corpus-sources.yaml| HE[hll-encode]
    CP -->|corpus-questions.yaml| HE
    HE -->|*.lp files| VA[validate]
    HE -->|encoding-mapping.yaml| VA
    CP -->|All artifacts| RE[review]
    HE -->|All artifacts| RE
    VA -->|All artifacts| RE
    RE -->|review-package.md| CO[commit]

Template System

Skills use templates from templates/:

Template Used By Purpose
corpus-report.md corpus-prep Human-readable source summary
encoding-report.md hll-encode Encoding summary
validation-report.md validate Test results summary
review-package.md review Complete review package

Best Practices

Starting an Encoding Session

  1. Verify prerequisites:
  2. Sefaria MCP available
  3. Mistaber engine working
  4. Clingo installed

  5. Start with corpus-prep:

    User: "Prepare corpus for YD 87:3"
    

  6. Review thoroughly at each checkpoint:

  7. Don't rush approvals
  8. Answer all questions
  9. Verify source accuracy

Handling Revisions

If issues are found after approval:

  1. Minor fixes: Request revision

    User: "Needs revision: Rule r_bb_dag_sakana should use sakana/1 not sakana/2"
    

  2. Major issues: May need to return to earlier phase

    User: "Reject - source interpretation is incorrect"
    

Resuming Sessions

User: "Resume encoding YD 87:3"

The skill system reads .mistaber-session.yaml and continues from the current phase.

Quick Reference

Trigger Phrase Summary

Skill Primary Triggers
corpus-prep "prepare corpus", "fetch sources", "start encoding", "begin encoding"
hll-encode "encode rules", "write HLL", "create ASP", "translate to HLL"
validate "validate", "run tests", "compile rules", "check rules"
review "review", "prepare review", "final review"
commit "commit", "finalize", "save encoding"
workflow-guide "how does encoding work", "workflow help"

Checkpoint Approval Phrases

Phrase Effect
"Approved" / "Looks good" / "Correct" Approve checkpoint
"Needs revision: [feedback]" Request changes
"Reject: [reason]" Cancel encoding