phase1-method-design

name

phase1-method-design

description

Main Agent orchestration skill for Phase 1 (Method Design). Load when the Experimental Setup Module has passed (EXPERIMENT_SETUP.md exists) and Phase 1 has not yet produced a confirmed method spec, or when re-entering Phase 1 after an upstream revision (IDEA.md or EXPERIMENT_SETUP.md changed). Drives the Architect → Confirmer loop with round cap and convergence detection. The Main Agent only monitors — it does not inject its own judgment between rounds. Hands off to Phase 2 (Small-Scale Verification).

Phase 1 — Method Design Pipeline

You are the Main Agent running Phase 1. Your role is to invoke the Architect, relay its spec to the Confirmer, relay the Confirmer's findings back to the Architect, and detect when the loop has converged. You do not design, revise, or evaluate the spec yourself.

Transparent-relay rule. Between the Architect and the Confirmer you are a pipe, not a participant. Do not rewrite the Confirmer's findings when handing them to the Architect; do not add your own observations to either side; do not pre-filter findings. The two agents interact through you because Claude Code subagents cannot invoke each other — not because you are editorializing.

Your only active decisions are: detecting ALL CLEAR, counting rounds, forcing exit at the cap, and escalating to the user on deadlock.

---

Entry — Identify the case

Before Round 1, determine which entry case applies:

Case A — Fresh entry from the Experimental Setup Module Loop 2 has passed, EXPERIMENT_SETUP.md exists, and no method_design_v*.md files exist in phase_1_design/. This is the normal path.

Case B — Re-entry after upstream revision A prior Phase 1 run existed, but IDEA.md or EXPERIMENT_SETUP.md has since been revised (typically because Phase 2 surfaced a Loop 3b — method redesign — that the user chose to route back here). Previous method_design_v*.md and confirmer_report_*.md files exist. Treat this as a full restart: archive the prior files to _logs/ (or let them stay — the round counter below will simply start from the highest existing {n}+1 to preserve history).

---

Round structure

Execute Steps 1–5 in order. Do not skip steps.

Step 1 — Confirm preconditions

Verify:

• phase_0_ideation/ideas/IDEA.md exists.
• phase_0_ideation/experiment_setup/EXPERIMENT_SETUP.md exists.
• The most recent Experimental Setup Module exit check passed, OR an acceptance note was placed at the top of EXPERIMENT_SETUP.md (user chose "accept and proceed" despite known gaps).

If any precondition is missing, halt and report to the user. Do not invoke the Architect.

Also create the output directory if it does not exist: phase_1_design/.

Determine the starting round number:

• Case A: round counter starts at 1 (initial spec → method_design_v1.md → confirmer_report_1.md).
• Case B: count existing method_design_v*.md files; start at the next unused index.

Step 2 — Invoke the Architect (initial spec — Round 1 only)

On Round 1 only, invoke the method-architect agent with this brief:

Produce the initial method design for the idea at phase_0_ideation/ideas/IDEA.md
with experimental setup at phase_0_ideation/experiment_setup/EXPERIMENT_SETUP.md.
This is Round 1 (initial spec). No revision note required.
Output: phase_1_design/method_design_v{n}.md   (use n = [starting round])

Wait for the Architect to finish. Confirm the file exists at the expected path before continuing.

On Round 2+, skip this step — the Architect revision invocation happens in Step 5 (after the Confirmer flags findings).

Step 3 — Invoke the Confirmer

Invoke the spec-confirmer agent with this brief:

Round 1 brief:

Review the method spec at phase_1_design/method_design_v{n}.md.
Anchors: phase_0_ideation/ideas/IDEA.md and phase_0_ideation/experiment_setup/EXPERIMENT_SETUP.md.
Round: {n} (initial review).
Output: phase_1_design/confirmer_report_{n}.md

Round 2+ brief:

Review the revised method spec at phase_1_design/method_design_v{n}.md.
Anchors: phase_0_ideation/ideas/IDEA.md and phase_0_ideation/experiment_setup/EXPERIMENT_SETUP.md.
Previous confirmer report: phase_1_design/confirmer_report_{n-1}.md
Round: {n}.
Output: phase_1_design/confirmer_report_{n}.md

Wait for the Confirmer to finish. Confirm the report exists at the expected path before continuing.

Step 4 — Check convergence

Read the Verdict from phase_1_design/confirmer_report_{n}.md.

If Verdict is ALL CLEAR: Phase 1 has converged. Proceed to the exit check (see Step 6 below).

If Verdict is REVISIONS NEEDED: proceed to Step 5.

Do not inspect the individual findings to form your own opinion. The Confirmer's binary verdict is authoritative for loop control. The findings are for the Architect, not for you.

Step 5 — Invoke the Architect (revision round)

This is where the loop turns. Increment the round counter: n ← n + 1.

Invoke the method-architect agent with this brief:

Revise the method spec at phase_1_design/method_design_v{n-1}.md to address
the Confirmer's findings.
Anchors: phase_0_ideation/ideas/IDEA.md and phase_0_ideation/experiment_setup/EXPERIMENT_SETUP.md.
Confirmer report to address: phase_1_design/confirmer_report_{n-1}.md
Round: {n} (revision).
Output: phase_1_design/method_design_v{n}.md

Wait for the Architect. Confirm the new spec file exists. Return to Step 3 with the incremented round counter.

Step 6 — Exit check and handoff

When Step 4 detected ALL CLEAR, run the phase1-exit-check skill against the latest method_design_v{n}.md and confirmer_report_{n}.md.

If PASS: Phase 1 is complete. Surface the result to the user:

Phase 1 (Method Design) — PASSED

Confirmed method spec: phase_1_design/method_design_v{n}.md
Convergence: Round {n} (ALL CLEAR from the Confirmer)

Module summary:
- [module names from the spec]

Known risks carried into Phase 2:
[list from the spec's Risk Flags section, or "none flagged"]

Deviations from EXPERIMENT_SETUP.md:
[list from the spec's Deviations subsection, or "none"]

The next step is Phase 2: Small-Scale Verification.

Then proceed to handoff (see "Handoff to Phase 2" below).

If NEEDS-REWORK: the exit check flagged a structural problem the Confirmer missed (rare — the Confirmer is the semantic check, the exit check is structural). Surface to the user with the exit-check's findings and the standard escalation menu (see "On deadlock" below).

---

Loop control

Hard cap: 3 Confirmer rounds.

A "round" is one Confirmer invocation. Round 1 reviews the initial Architect spec; Round 2+ reviews each Architect revision. The round counter is cumulative for this Phase 1 entry — it does not reset when an Architect revision is requested.

If Round 3's Confirmer Verdict is REVISIONS NEEDED, do not begin Round 4 automatically. Escalate (see "On deadlock").

---

On deadlock

When the round cap is hit without convergence, surface the situation to the user with the latest Confirmer findings:

Phase 1 (Method Design) — Round 3 — REVISIONS STILL NEEDED

The Architect ↔ Confirmer loop has run 3 rounds without converging.

Latest Confirmer findings:
[Summarize the open findings from confirmer_report_3.md — category + count +
the 1–2 most consequential specific findings. Do not paraphrase into your own
assessment.]

Recommended interpretation (from the pattern of findings):
[Choose one and state it plainly:
 (i) the IDEA underspecifies something the Architect cannot invent — route to
     Loop 1 to revise IDEA.md;
 (ii) the EXPERIMENT_SETUP is missing a resource the method requires — route
      to the Experimental Setup Module to supplement;
 (iii) the spec has a genuinely hard design question the Architect cannot
       resolve inside one loop — user input is needed on the design itself;
 (iv) the Confirmer is raising pedantic findings that do not warrant another
      round — accept the current spec with the open findings noted.]

Options:
(a) Continue — allow one more Architect ↔ Confirmer round.
(b) Accept current spec — proceed to handoff despite open findings. I will
    record the accepted open findings at the top of method_design_v{n}.md so
    Phase 2 is aware.
(c) Retreat to Loop 1 — IDEA.md needs to change before the design can converge.
    I will invoke loop1-ideation-pipeline (Case C) with the Confirmer's
    findings as the standing requirement.
(d) Retreat to the Experimental Setup Module — EXPERIMENT_SETUP.md is
    insufficient. I will invoke experiment-setup-pipeline (Case B) with a
    scout supplement request scoped to the missing resource.
(e) Abandon — discard the current Phase 1 attempt.

Wait for user instruction. Do not auto-route.

On user response:

• (a) Continue: run one more round, then re-escalate if still not converged. Do not loop indefinitely — escalate again at the next REVISIONS NEEDED.
• (b) Accept current spec: write a short note at the top of the latest method_design_v{n}.md — "Accepted with open findings on [date]; see confirmer_report_{n}.md — user explicitly overrode non-convergence." Run the exit check; if it still NEEDS-REWORK for structural reasons, surface those separately. Proceed to handoff.
• (c) Retreat to Loop 1: invoke loop1-ideation-pipeline with Case C entry and the Confirmer findings as the standing requirement. When Loop 1 + Loop 2
  • Experimental Setup complete, re-enter this pipeline as Case B.
• (d) Retreat to Experimental Setup: invoke experiment-setup-pipeline Case B with the missing-resource gap as the supplement brief. When the re-audit passes, re-enter this pipeline as Case B.
• (e) Abandon: stop. Record the abandonment reason in phase_1_design/abandon_note.md.

---

Handoff to Phase 2

Phase 2 (Small-Scale Verification) is not yet built. For now, simply announce:

Phase 1 is complete. The Coder, Fidelity Checker, and Phase 2 pipeline are not
yet implemented. To resume, the next session should pick up at Phase 2
implementation once those are in place.

Phase 1 outputs ready for Phase 2:
- phase_1_design/method_design_v{n}.md   (confirmed spec — the spec the Coder
  implements)
- phase_1_design/confirmer_report_{n}.md (the ALL CLEAR verdict, for traceability)
- phase_0_ideation/ideas/IDEA.md         (unchanged since Phase 1 entry)
- phase_0_ideation/experiment_setup/EXPERIMENT_SETUP.md (unchanged)

The phase_1_design/PHASE_1_SUMMARY.md file (per file_organization_structure.md) is the natural next artifact — when Phase 2 is built, that skill writes it on entry. Do not write it here.

---

Files produced by this pipeline

File	Written by	Notes
phase_1_design/method_design_v{n}.md	Architect	One per round; v{n} is the spec reviewed by confirmer_report_{n}.md. Never overwritten — history is preserved so rollback (Loop 3b) can compare against earlier designs.
phase_1_design/confirmer_report_{n}.md	Confirmer	One per round, matched to the same {n} as the spec version. Never overwritten.
phase_1_design/abandon_note.md	Main Agent	Only written if the user chooses (e) Abandon.

Why files are versioned and never overwritten. Unlike EXPERIMENT_SETUP.md (single living audit), Phase 1 produces a design history that Loop 3b rollbacks need to consult — "did we already try this design and discard it, or is this genuinely a new idea?" is a question only a preserved history can answer. The final confirmed spec is whichever method_design_v{n}.md corresponds to the ALL CLEAR Confirmer round.

---

Interaction with other loops

• From the Experimental Setup Module: enter as Case A after Experimental Setup PASS or after user accepted gaps.
• To Loop 1: only on user choice (c) at deadlock — a standing-requirement retreat because IDEA.md underspecifies.
• To the Experimental Setup Module: only on user choice (d) at deadlock — a supplement because EXPERIMENT_SETUP.md is insufficient for the method.
• To Phase 2: via handoff once the confirmed spec and an ALL CLEAR Confirmer report exist.
• Re-entry (Case B): triggered later by Phase 2's Loop 3b (when Phase 2 is built). Not triggered inside this module.