Methodology Intake

Methodology Intake

Use this as a read-only classifier before turning external methodology sources into repo artifacts. Do not create or modify skills, specs, ADRs, issues, or glossary terms during intake unless the user explicitly asks for the follow-up work after reviewing the report.

Inputs

• External methodology source: article, repository, X thread, paper, tool list,

framework, workflow, or public discussion.

• Optional captured input note from capture-input-note.
• Optional user goal: learn, critique, adapt, implement, reject, or park.

Workflow

1. Read the source, captured input note, or provided excerpt closely enough to

classify it.

1. Check current repo language in CONTEXT.md and relevant roadmap/spec docs.
2. Produce one Markdown methodology intake report.
3. Assign exactly one primary intake destination.
4. Assign exactly one retention status.
5. Explain why the other destinations were not selected.
6. If implementation is recommended, name the minimum vertical slice and

verification.

1. If retention is parked or revisit-on-trigger, record a dated one-line

verdict (source, destination, retention, trigger) in the repo's intake ledger — in this repo, docs/intake.md. The verdict line is part of the intake output, not follow-up work; do not write anything else there.

Destinations

Use exactly one primary destination:

• Rule: A judgment principle that can be absorbed into an existing workflow.
• Skill: A repeatable workflow with trigger criteria, inputs, steps, outputs,

verification, stop conditions, and repo fit.

• Context term: Stable project language that will be reused across workflows,

specs, or skills and affects classification, acceptance, or boundaries.

• ADR: A hard-to-reverse, surprising-without-context repo decision that comes

from a real trade-off surfaced during intake.

• Spec: Source material that should define or change cross-workflow

structure, process, contracts, artifact policy, or acceptance rules.

• Issue: A tracked, scoped, independently verifiable work item.
• No-op: No formal workflow artifact should be created or updated.

Retention Status

Use exactly one retention status:

• discard: Do not retain the source. Use for duplicates, low-signal material,

irrelevant material, or unacceptable license / attribution risk.

• parked: Retain as background material, but do not process further now.
• revisit-on-trigger: Retain and define the concrete trigger that should make

someone re-evaluate it.

No-op does not mean delete or forget. A no-op source may still be parked or revisited later.

Retention decisions must survive the conversation: parked and revisit-on-trigger require a verdict line in the repo's intake ledger (workflow step 8); discard requires none. discard also means no revisit trigger — if the verdict names a concrete condition for re-evaluating the source, the retention is revisit-on-trigger, not discard. Retention keeps the pointer and verdict, not a copy of the source.

Skill Admission Gate

Classify a source as Skill only when it shows a repeatable workflow:

• trigger: when to use it
• input: required source material or context
• workflow: repeatable steps, not just advice or a tool list
• output: artifact or decision produced
• verification: how the result is judged acceptable
• stop condition: when to stop exploring or iterating
• repo fit: how it belongs in the current skills/specs/workflow system

If these signals are missing, prefer Rule, Spec, Issue, Context term, or No-op.

Relationship To Capture Input Note

capture-input-note ingests an external source as a redacted work-log inbox note when source access or provenance work is needed. methodology-intake classifies whether that source should change repo artifacts.

Use a captured input note as the preferred input when the source is authenticated, private, long, noisy, likely to disappear, or needs redaction. Do not create persistent captures inside methodology intake; run capture-input-note first when capture is needed.

Tool-List-Only Sources

A tool-list-only source primarily lists tools, SDKs, repositories, or frameworks without enough workflow, verification, or stop-condition detail. Do not promote it directly into a skill.

Typical result:

destination: No-op
retention: parked

Use Rule instead of No-op only when the source clearly supports a reusable judgment principle for an existing workflow.

Report Template

# Methodology Intake Report

## Source

- title:
- url_or_path:
- source_type: article | repo | X thread | paper | tool list | workflow | other
- license_or_attribution_status:

## Classification

- destination: Rule | Skill | Context term | ADR | Spec | Issue | No-op
- retention: discard | parked | revisit-on-trigger
- confidence: low | medium | high

## Rationale

- why this destination:
- why not the other destinations:

## Evidence

- reusable workflow signals:
- missing signals:
- source excerpts or references:

## Risks

- adoption risk:
- license / attribution risk:
- repo fit risk:

## Recommended Next Step

- action:
- revisit_trigger: (n/a when retention is discard)
- minimum vertical slice, if any:
- verification:

Acceptance Examples

Use these examples to sanity-check the classification behavior:

• Tool-list-only source: classify as No-op or Rule; set retention to

parked or revisit-on-trigger; do not recommend creating a skill directly.

• Visual prompt reconstruction workflow: classify as candidate Skill when it

has a generation/evaluation/refinement loop; identify guardrails such as authorization, copyright, evaluator, max iterations, and stop condition; name a minimum vertical slice before implementation.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: justinmchoi
• Source: justinmchoi/ai-toolkit
• License: MIT

Install and usage instructions live in the source repository linked above.