CtrlK
BlogDocsLog inGet started
Tessl Logo

lab-session

Scaffold a JM Labs Lab session: the 4 canonical files (notas, hipotesis, referencias, decision) per protocol P08, missing-only and never overwriting local edits.

SKILL.md
Quality
Evals
Security

Lab Session

Materializes one Lab session as the four canonical files of protocol P08: notas.md, hipotesis.md, referencias.md, decision.md. [DOC] One run creates one session folder; reruns are additive (missing-only). [INFERENCE]

When to Use

  • User asks to start / create a Lab, an experiment, or a P08 session. [DOC]
  • A loose idea or hypothesis needs a structured place to live before it becomes a project. [INFERENCE]
  • Use when the four-file scaffold is the goal. NOT for: editing an existing session's content (use Read+Edit directly), producing client deliverables (wrong brand — see Anti-Scope), or running the experiment itself. [INFERENCE]

Inputs

InputRequiredDefault if absent
Session topic / objectiveyesSTOP and ask — terminal gap [ASSUMPTION]
Session slug (kebab-case)noderive from topic [INFERENCE]
Target Lab root pathnocurrent working dir [ASSUMPTION]
Initial hypothesisnohipotesis.md left as a prompt stub [INFERENCE]
Seed referencesnoempty referencias.md skeleton [INFERENCE]

Empty objective is a critical gap: do not invent a topic — ask. [INFERENCE]

Outputs

A <lab-root>/<slug>/ folder containing exactly four files:

  • notas.md — running log, observations, dated entries.
  • hipotesis.md — the claim under test, framed falsifiably. [INFERENCE]
  • referencias.md — sources, links, prior art; each entry carries a tag.
  • decision.md — the verdict (keep / pivot / kill) once the Lab closes; starts empty/{POR_CONFIRMAR}. [INFERENCE]

Plus a one-line console summary of created vs. skipped (pre-existing) files.

Procedure

1. Discover

Resolve slug and target path. Bash a check for an existing <slug>/ folder and any of the four files. [INFERENCE]

2. Plan

Classify each of the four target paths as CREATE (absent) or SKIP (present). Never plan an overwrite without an explicit --force from the user. [DOC]

3. Execute

Write only the CREATE files using the canonical skeleton per file (see Outputs). Touch nothing classified SKIP. Keep folder scope to the one session.

4. Validate

Re-list the folder; confirm all four files exist and CREATE files are non-empty. Emit the created/skipped summary.

Validation Gate (acceptance criteria)

  • Exactly the four canonical files present after run. [DOC]
  • No pre-existing file was modified (byte-compare or mtime on SKIPs). [INFERENCE]
  • hipotesis.md states a falsifiable claim or an explicit stub. [INFERENCE]
  • Every referencias.md entry and every non-obvious note carries one Alfa-core tag ([DOC]/[CONFIG]/[CODE]/[INFERENCE]/[ASSUMPTION]), one spelling throughout. [DOC]
  • Summary line reports created vs skipped counts.

Fail any box → fix before declaring done; do not green-light partial scaffolds. Machine-readable gate + per-file skeletons: assets/ (see assets/README.md). [DOC]

Edge Cases

  • Folder exists, all four files present: report "already scaffolded", change nothing, exit 0. [INFERENCE]
  • Partial scaffold (some files present): create only the missing ones. [INFERENCE]
  • Slug collision with a different topic: surface the conflict, ask before reusing the folder. [INFERENCE]
  • Non-writable target path: stop with the path error, do not fall back to a surprise location. [ASSUMPTION]
  • Empty objective: terminal gap — ask, never auto-fill. [INFERENCE]

Anti-Scope

  • Not a deliverable generator and not a brand-rendered document — Lab files are internal JM Labs scratch, never client-facing. [INFERENCE]
  • Single-brand: JM Labs only. Do NOT emit MetodologIA or MetodologIA framing, palettes, or footers here. [DOC]
  • Never introduce green as a success signal in any sample content. [DOC]
  • Does not run, score, or conclude the experiment — only the verdict file is scaffolded; the human writes decision.md. [INFERENCE]

Self-Correction Triggers

  • About to overwrite a present file → STOP, reclassify as SKIP. [INFERENCE]
  • Tempted to fabricate a topic from thin context → STOP, ask. [INFERENCE]
  • More or fewer than four files in the plan → the P08 contract is broken; fix. [DOC]
  • Mixed tag spellings or families → normalize to one before delivery. [DOC]

Update-Safety Notes

  • All four files are missing-only by default. [DOC]
  • --force regenerates a file only after the user reviews the diff. [DOC]
  • Local/experimental variants live under .local/ and are never overwritten. [INFERENCE]

Related Skills

  • workspace-governance — where the Lab folder fits in the workspace tree.
  • workflow-forge — promoting a kept Lab into a real workflow.
  • quality-guardian — independent check of the validation gate.

Evidence Convention

Alfa core set (kit-facing), EN spelling, one tag per claim: [CODE] / [CONFIG] / [DOC] / [INFERENCE] / [ASSUMPTION]. Canon and homologation: references/verification-tags.md. [DOC]

Repository
JaviMontano/jm-adk-beta
Last updated
First committed

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.