feat: add experiment-spec skill for research projects #1

Merged
mathias merged 1 commits from feat/experiment-spec-skill into main 2026-06-22 17:33:26 +00:00
Owner

What this does

Adds the experiment-spec skill to the shared skills repo. This is the research analogue of feature-spec — same discipline (falsifiable hypothesis, quantitative criteria, explicit out-of-scope, null result protocol), adapted vocabulary for scientific/ML experiments.

Why now

The jepa-fx-risk project is the first research project in the stack. feature-spec is designed around product features and user stories. Research experiments need: hypothesis framing, baseline comparison, null result protocol, training cutoff discipline. These don't map cleanly onto feature-spec.

Key design decisions

  • Iron Laws mirror feature-spec (quantitative criteria, mandatory out-of-scope) but add: baseline is mandatory, null result protocol is mandatory, training cutoff is sacred
  • Brain MCP integration: brain_query wing=X hall=decisions + hall=failures at spec start; brain_write to hypotheses/ after approval
  • Cross-references to feature-spec (for component implementation within a phase), grill-me (stress-test before running), tdd (criteria → tests), session-retrospective (after phase)
  • Worked example uses Phase 0 SSL feasibility spec from jepa-fx-risk — concrete and immediately useful

Relates to

  • mathias/jepa-fx-risk PR #1 — first consumer of this skill
## What this does Adds the `experiment-spec` skill to the shared skills repo. This is the research analogue of `feature-spec` — same discipline (falsifiable hypothesis, quantitative criteria, explicit out-of-scope, null result protocol), adapted vocabulary for scientific/ML experiments. ## Why now The `jepa-fx-risk` project is the first research project in the stack. `feature-spec` is designed around product features and user stories. Research experiments need: hypothesis framing, baseline comparison, null result protocol, training cutoff discipline. These don't map cleanly onto `feature-spec`. ## Key design decisions - Iron Laws mirror `feature-spec` (quantitative criteria, mandatory out-of-scope) but add: **baseline is mandatory**, **null result protocol is mandatory**, **training cutoff is sacred** - Brain MCP integration: `brain_query wing=X hall=decisions` + `hall=failures` at spec start; `brain_write` to `hypotheses/` after approval - Cross-references to `feature-spec` (for component implementation within a phase), `grill-me` (stress-test before running), `tdd` (criteria → tests), `session-retrospective` (after phase) - Worked example uses Phase 0 SSL feasibility spec from `jepa-fx-risk` — concrete and immediately useful ## Relates to - `mathias/jepa-fx-risk` PR #1 — first consumer of this skill
mathias added 1 commit 2026-05-27 22:01:00 +00:00
mathias merged commit 5ffa397029 into main 2026-06-22 17:33:26 +00:00
Sign in to join this conversation.
No Reviewers
No labels
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: mathias/skills#1