loop2-stress-test

name	loop2-stress-test
description	Main Agent orchestration skill for Loop 2 (stress testing). Load when Loop 1 has passed and the idea is ready for adversarial evaluation, or when re-entering Loop 2 after a Loop 1 retreat fixed the Tester's attack points. Drives the Tester → exit-check → retreat-or-proceed cycle. Controls loop termination and user notification.

Loop 2 — Stress Test Pipeline

You are the Main Agent running the stress testing loop. Your role is to invoke the Tester, evaluate its findings via the exit check, and decide whether to proceed or retreat. You do not perform the stress testing yourself.

Do not attempt to fix attack points. If the idea needs revision, that is the Refiner's job via Loop 1 retreat.

---

Entry — Identify the case

Before Round 1, determine which entry case applies:

Case A — Fresh entry from Loop 1 The idea has just passed Loop 1's exit check for the first time (or after a full restart). No previous Tester reports exist for this idea version.

Case B — Re-entry after Loop 1 retreat The idea was revised in Loop 1 to address Tester attack points from a previous Loop 2 round. Previous Tester reports exist in phase_0_ideation/stress_test/.

---

Round structure

Execute Steps 1–4 in order. Do not skip steps.

Step 1 — Prepare context

Read phase_0_ideation/ideas/IDEA.md to confirm it is the version you expect to test.

For Case B: also read the most recent tester_report_*.md in phase_0_ideation/stress_test/ to identify which attack points prompted the Loop 1 retreat. You will pass this context to the Tester so it can assess whether those points were addressed.

Step 2 — Invoke the Tester

Invoke the Tester agent (idea-pressure-tester) with a brief:

Case A brief:

Stress-test the idea at phase_0_ideation/ideas/IDEA.md.
This is a fresh evaluation — no prior Tester reports exist.
Output tester report to: phase_0_ideation/stress_test/tester_report_{n}.md
Output adversarial literature to: phase_0_ideation/stress_test/tester_literature_{n}.md

Case B brief:

Re-test the idea at phase_0_ideation/ideas/IDEA.md.
The idea was revised to address attack points from the previous round.
Previous tester report: phase_0_ideation/stress_test/tester_report_{n-1}.md
Assess whether previous Fatal/Serious attacks have been addressed.
Look for new weaknesses introduced by the revision.
Output tester report to: phase_0_ideation/stress_test/tester_report_{n}.md
Output adversarial literature to: phase_0_ideation/stress_test/tester_literature_{n}.md

where {n} is the next available version number — count existing tester_report_*.md files in phase_0_ideation/stress_test/ and increment by one.

Wait for the Tester to finish. Confirm both output files exist before continuing.

Step 3 — Evaluate

Run the loop2-exit-check skill against the latest Tester report.

If all criteria pass: Loop 2 is complete. The idea has survived stress testing. Announce to the user:

Loop 2 (stress testing) — PASSED

The idea has survived adversarial evaluation. Summary:
[Brief summary of what was tested and the result]

Remaining known risks (Serious-level, non-blocking):
[List from exit check, or "None"]

The idea is ready to proceed to the Experimental Setup Module.

Proceed to the next phase.

If any criteria fail: go to Step 4.

Step 4 — Retreat decision and user notification

When the exit check fails, the idea needs revision. This means retreating to Loop 1 with the Tester's attack points as direction.

Notify the user before retreating. Present the Tester's findings and your recommended action:

Loop 2 (stress testing) Round [n] — FAILED

The Tester identified the following issues:

[For each Fatal and Serious attack point, summarize in 2–3 sentences:
 - What the problem is
 - What evidence supports it
 - Its severity]

Failing exit criteria:
[List from exit check]

Recommended action: retreat to Loop 1 to address these attack points.

Options:
(a) Retreat — I will invoke Loop 1 with the Tester's attack points as direction.
    The Refiner will revise IDEA.md to address them.
(b) Retreat with guidance — same as (a), but you provide additional direction
    or constraints for the revision. Describe your guidance below.
(c) Override — proceed despite failing criteria. The idea will enter the
    Experimental Setup Module with known unresolved issues.
(d) Abandon — discard this idea and return to ideation.

Wait for user instruction. Do not auto-retreat.

On user response:

• (a) Retreat: invoke loop1-ideation-pipeline with Case C entry. The Tester's Fatal and Serious attack points become the standing requirements. When Loop 1 completes (passes exit check), return to Loop 2 as Case B.

• (b) Retreat with guidance: same as (a), but append the user's guidance to the standing requirements when invoking Loop 1.

• (c) Override: record the override decision and unresolved issues in phase_0_ideation/stress_test/override_note.md. Proceed to the next phase.

• (d) Abandon: stop. Do not proceed further. Record the abandonment reason.

---

Loop control

Hard cap: 3 rounds.

A "round" is one complete Tester invocation + exit check. The count does NOT reset when retreating to Loop 1 — the round counter is cumulative across the entire Loop 2 entry for this idea.

If Round 3 ends and the exit check still fails, do not begin Round 4. Surface the situation to the user:

Loop 2 has completed 3 rounds without passing the exit check.

Remaining issues:
[Fatal and Serious attacks still unaddressed]

Options:
(a) Continue for additional rounds — confirm to proceed.
(b) Override — proceed to Experimental Setup Module with known issues.
(c) Redirect the idea — describe a new direction and I will restart from Loop 1.
(d) Abandon — discard this idea.

Wait for user instruction before acting.

---

Files produced by this loop

File	Written by	Notes
phase_0_ideation/stress_test/tester_report_{n}.md	Tester	One per round, globally numbered, never overwritten
phase_0_ideation/stress_test/tester_literature_{n}.md	Tester	Matches tester_report version number
phase_0_ideation/stress_test/override_note.md	Main Agent	Only written on user override (option c)

Why reports are numbered and never overwritten: Each round's Tester report captures the state of the idea at that point. Re-test rounds need access to previous reports to assess whether attacks were addressed. The adversarial literature reports accumulate distinct search results that may be reused by the Experimental Setup Module.

---

Interaction with Loop 1

When retreating to Loop 1:

• Loop 1 runs with Case C entry (Tester attack points as standing requirements)
• Loop 1 has its own 3-round hard cap — if it cannot resolve the attacks within 3 rounds, it escalates to the user (per Loop 1's own control logic)
• When Loop 1 completes (passes its exit check), control returns here as Case B
• The Loop 2 round counter increments — the retreat does not reset it