| name | stress-test |
| description | Use when a design, plan, or decision needs adversarial scrutiny before proceeding. Interrogates every branch of the decision tree, providing recommended answers and forcing explicit agreement or pushback. Triggers on "grill me", "stress test this", "poke holes", "challenge this design", or when brainstorming/writing-plans suggests review. |
Stress Test: Adversarial Design Interrogation
Announce at start: "I'm using the stress-test skill to interrogate this design."
Purpose
Stress-test a design, plan, or decision by walking down every branch of the decision tree. For each question, provide your own recommended answer — don't just ask, propose. This forces the user to either agree explicitly or articulate why their approach is better.
This is NOT brainstorming (which creates designs) or verification (which checks implementations). This is the gap between them: "Is this design actually solid before we commit to building it?"
When to Invoke
| Trigger | Context |
|---|
| After brainstorming | Stress-test the design spec before writing a plan |
| After writing-plans | Stress-test the plan before execution begins |
| User says "grill me" | On-demand for any document, decision, or approach |
| Before a major architectural decision | Ensure alternatives were genuinely considered |
The Process
bd create "Stress-test: <topic>" -t chore -p 2 \
--description="Adversarial review of <artifact>. Branches to interrogate: <count>"
bd update <id> --claim
Phase 1: Understand the Target
Read the design, plan, or decision document thoroughly. If no document exists, ask the user to describe the approach. Explore the codebase for context — answer your own questions from code when possible rather than asking the user.
Restore point (Mode A only): If the target artifact has uncommitted changes, commit or stash them before starting — this preserves a clean restore point before inline edits begin. In the normal flow (brainstorming → stress-test), the artifact is already committed.
Done when: the target is understood well enough to enumerate its decision branches, and (Mode A) the artifact sits at a clean restore point.
Phase 2: Map the Decision Tree
Identify every decision branch in the target:
- Architecture choices (why X over Y?)
- Assumptions (what breaks if this is wrong?)
- Dependencies (what happens if this changes?)
- Edge cases (what about when Z happens?)
- Scale (does this work at 10x? 100x?)
- Failure modes (what's the worst case?)
- Alternatives not considered (what about approach W?)
- Security & risk (mandatory branch): does any branch take a shortcut, descope a requirement, or accept material risk? Does anything weaken or bypass a security control, or introduce a vulnerability? Per the Production-Grade Doctrine, a design that does fails the stress test by default. (If the design has no security surface, resolve this branch as "no security surface — N/A" — do not fabricate a finding.)
Done when: every branch category above has been checked against the target, including the mandatory Security & risk branch.
Phase 3: Interrogate One Branch at a Time
For each branch, present your question and recommendation as text, then use your structured question tool for the response.
Per-branch flow:
- Present the question + recommendation as text in the message body (reasoning needs room to breathe)
- Immediately follow with a structured question (content below; shape shown in Claude Code schema — adapt to your tool):
{
"questions": [{
"question": "<1-sentence summary of the branch being interrogated>",
"header": "Stress test",
"options": [
{"label": "Agree", "description": "Accept the recommendation and move to the next branch"},
{"label": "Disagree", "description": "I have a different view — let me explain"},
{"label": "Discuss further", "description": "I want to explore this branch more before deciding"}
],
"multiSelect": false
}]
}
Response handling:
- Agree — Mark branch resolved, emit status line, advance to next branch
- Disagree — Ask "What's your alternative?" as text (open-ended — disagreements need space). Iterate until the branch resolves, then re-ask the same 3-option structured question on the revised position.
- Discuss further — Explore deeper (code, docs, implications), present updated analysis, then re-ask the same structured question
Branch tracking: After each branch resolves, emit a status line:
✓ Resolved: 3/7 branches (2 agreed, 1 modified)
Remaining: Error handling, Scale, Rollback, Testing strategy
Rules:
- One branch at a time — never batch. Wait for the user's response on each branch before presenting the next; surfacing several at once is bewildering and dilutes the recommendation each one deserves.
- Always state your recommendation in the message body BEFORE the structured question — the recommendation is the substance; the click is just the gate
- If you can answer by exploring the codebase, do that instead of asking
- When the user agrees, move on. When they push back, explore deeper.
Done when: every mapped branch is marked resolved and the status line reads N/N.
Phase 4: Document Findings
After all branches are resolved, write the findings. The output mode depends on context.
Mode detection:
- Mode A applies when: the stress-test was invoked by brainstorming or writing-plans (caller passes the artifact path), OR the user explicitly points at a
.internal/specs/ or .internal/plans/ file.
- Mode B applies for everything else: user-initiated "grill me" with no artifact, stress-testing a conversation or decision, or targeting documents that shouldn't be edited inline (README, CLAUDE.md, etc.).
- When ambiguous: Use your structured question tool to ask:
{
"questions": [{
"question": "I see `<file>`. Should I edit it inline with findings, or produce a separate stress-test report?",
"header": "Output mode",
"options": [
{"label": "Edit inline (Mode A)", "description": "Apply changes directly to the source document and append a results summary"},
{"label": "Separate report (Mode B)", "description": "Write findings to .internal/stress-tests/ without modifying the source"}
],
"multiSelect": false
}]
}
Mode A — Existing artifact (spec, plan, design doc in .internal/):
- Edit the source artifact directly when a branch changes the design.
- At the end, append a
## Stress Test Results section at the bottom of the source document:
## Stress Test Results: <topic>
### Resolved Decisions
- [Decision 1]: [Resolution and rationale]
- [Decision 2]: [Resolution and rationale]
### Changes Made
- [Any modifications to the original design/plan]
### Deferred / Parking Lot
- [Items explicitly deferred for later]
### Confidence Assessment
- Overall: High/Medium/Low
- Areas of concern: [Any remaining worries]
Alternatively, record as a bd note on the parent bead if the source doc shouldn't be modified further.
Mode B — Standalone stress test (no existing artifact):
- Create
.internal/stress-tests/YYYY-MM-DD-<topic>.md with the full findings template above.
- Open in user's editor for review:
User's preferred editor: !echo ${VISUAL:-${EDITOR:-not-configured}}
if [ -n "$VISUAL" ]; then
"$VISUAL" "<findings-file-path>"
elif [ -n "$EDITOR" ]; then
"$EDITOR" "<findings-file-path>"
elif command -v open >/dev/null 2>&1; then
open "<findings-file-path>"
else
xdg-open "<findings-file-path>" 2>/dev/null
fi
Done when: findings exist in the applicable form — Mode A's Results section appended (or a bd note recorded), or Mode B's report file created.
Phase 4.5: Reflexion Self-Review
After documenting findings, run a single self-critique pass. This is internal reasoning — not shown to the user. Only the consequences (new or re-opened branches) are visible.
Self-critique questions:
- Coverage: Compare branches mapped in Phase 2 against branches actually interrogated. List any that were skipped or merged, with justification.
- Depth: Did I challenge the design, or just confirm what was already there? Were my recommendations genuinely independent, or did I echo the existing approach? Did I accept any "it's fine" answers without specific reasoning?
- Missed angles: What failure modes, alternatives, or assumptions did I NOT explore?
Resolution:
- Coverage gaps (branches mapped but not interrogated) → go back and interrogate them
- Depth issues (sycophantic agreement, rubber-stamping) → re-interrogate the weakest branches with harder questions
- Missed angles (genuinely new branches) → add to the map and interrogate them
Visible consequence: If reflexion adds or re-opens branches, emit an update via the branch tracking status line:
✓ Resolved: 7/7 branches (5 agreed, 2 modified)
[Reflexion added 2 new branches]
✓ Resolved: 9/9 branches (7 agreed, 2 modified)
Termination rule: Reflexion runs exactly once. One self-critique pass, address what it finds, then proceed to Phase 5. No recursive reflexion.
Done when: the one self-critique pass has run and its coverage, depth, and missed-angle findings are resolved.
Phase 5: Close
⚠️ Run the open command as a standalone Bash call — never chain it after bd commands in the same invocation (e.g., bd close <id> && open file.md). The combination hangs.
bd close <id> --reason "Stress-test complete: N branches resolved, M changes made, confidence: <level>"
After the work is settled, present the Capture gate — mandatory every time; Skip is the default (most work leaves nothing worth keeping):
{
"questions": [{
"question": "Worth keeping anything from this?",
"header": "Capture",
"options": [
{"label": "Skip", "description": "Nothing here outlasts the work itself (usually the case)"},
{"label": "Record the decision", "description": "Pick this if the choice is hard to undo, non-obvious in hindsight, and had real trade-offs — so future-you knows why"},
{"label": "Remember the lesson", "description": "A specific, evidence-backed lesson worth reusing in later sessions"},
{"label": "Both", "description": "A lasting decision and a lesson worth reusing"}
],
"multiSelect": false
}]
}
Route on the answer. Record the decision / Both → this writes an ADR, so first confirm it clears the bar (hard-to-reverse AND surprising-without-context AND genuine trade-off); if it doesn't, say so and capture it as a memory instead (the lighter record) — unless the user confirms they want the full ADR. Write the ADR (docs/decisions/ADR-NNNN-<kebab>.md, sections Context/Decision/Rationale/Consequences, update docs/decisions/INDEX.md), then file a type=decision knowledge-bead so the decision stays retrievable: printf '%s' "<distilled 0.5-2.5KB decision summary — context, decision, consequences>" | bd create "<one-line summary>" -t decision -l kb,adr-process,<topic> --defer 2099-01-01 --metadata "$(jq -nc --arg d "<ADR-path>" '{doc:$d}')" --body-file - --silent (run the secret/PII scan on the summary first — flag for removal, never write a secret into a bead). Remember the lesson / Both → bd remember "<kind>: <durable, evidence-backed insight>". Skip → nothing.
Done when: bd close has run with evidence and the Capture gate has been presented and routed (including Skip).
Anti-Rationalization
| Shortcut | Reality |
|---|
| "I asked 3 questions, that's enough" | Cover ALL major branches — count the decision tree, not the questions |
| "This is a simple project" | Simple projects have the most unexamined assumptions |
| "We already brainstormed this" | Brainstorming proposes; stress-testing challenges |
| "I don't want to slow things down" | Catching a flaw now saves 10x the time later |
| "They clicked Agree fast, so this is going well" | Speed ≠ depth — fast agreement might mean they're not reading your recommendation carefully. Don't reduce rigor. |
| "It's a reasonable trade-off" | Name the downside and its blast radius. A material-risk trade-off is surfaced to the user, never waved through — and a security regression is never acceptable. |
| "Security's out of scope for this design" | Security is in scope for every design. If a branch touches auth, data, input, or secrets, interrogate it. |
Red Flags
Never:
- Skip branches because they seem obvious
- Accept "it's fine" without specific reasoning
- Wave through a shortcut, a silent descope, a material-risk trade-off, or a security regression — these fail the stress test by default (Production-Grade Doctrine)
Always:
- Provide a recommended answer for every question
- Track resolved vs unresolved branches
- Produce a written findings summary
- Create and close a bead with evidence
Integration
Called by:
- brainstorming — offered at the spec-review gate every time; runs between design approval and writing-plans
- writing-plans — offered at the plan-review gate every time; runs between plan approval and execution
- Any workflow where a decision needs adversarial scrutiny
Pairs with:
- brainstorming — stress-test validates what brainstorming produced
- writing-plans — stress-test validates what the plan proposes
- verification-before-completion — stress-test for designs, verification for implementations
