debate
Skill8 starsChanged 6 months agoMITView on GitHub
Compare significant design approaches before implementation: generate three real options, use a compact matrix and separated hostile review, test edge cases, and converge in two to three rounds. Not for brainstorming, implementation, or code review.
26 other projects use a version of this.
Adopted by
Most recently taken up by
- Lu1sDV/skillsmdtheir own version, first seen Nov 4, 2026
- yoshibtc/multi-model-debate-skilltheir own version, first seen Oct 15, 2026
- phant0um/claude-skillstheir own version, first seen Oct 15, 2026
- helderberto/agent-skillstheir own version, first seen Oct 13, 2026
- rahulrajaram/rahulskillstheir own version, first seen Oct 13, 2026
- agent-sh/debatetheir own version, first seen Oct 13, 2026
Used with
Projects with it also use handoff, commit and 2 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
MicrosoftSkillhandoff
Compact the current conversation into a handoff document for another agent to pick up.
Also in 5 projects that use this
SentrySkillcommit
Create commit messages following Sentry conventions. Use when committing code changes, writing commit messages, or formatting git history. Follows conventional commits with Sentry-specific issue references.
Also in 3 projects that use this
GoogleSkillgrill-me
A relentless interview to sharpen a plan or design.
Also in 3 projects that use this
SentrySkillskill-creator
Create new agent skills following the Agent Skills specification. Use when asked to "create a skill", "add a new skill", "write a skill", "make a skill", "build a skill", or scaffold a new skill with SKILL.md. Guides through requirements, writing, registration, and verification.
Also in 3 projects that use this
Other versions26 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 26
- raine/consult-llm-mcp★ 148same name
- wolverin0/claude-skills★ 41same name
- helderberto/agent-skills★ 19same name
- ggombee/code-forge★ 18same name
- STRML/cc-debate★ 10same name
- Magnus-Gille/claude-skills★ 5same name
- agent-sh/debate★ 4same name
- HartreeWorks/research-skills★ 4same name
- ryancproduct/product-ai-toolkit★ 4same name
- Yafan-Yang/ai-trading★ 4same name
16 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository129 other files it gives its agents.
- AGENTS.mdAGENTS.md
- guidance/AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- guidance/CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
- blast-radius-checkSkill
- brainstormingSkill
- branch-flow-gateSkill
- branch-sync-gateSkill
- codebase-reconSkill
- codeowners-drift-auditSkill
- code-review-batterySkill
- cognitive-complexity-refactoringSkill
- debug-conductorSkill
- dispatching-parallel-agentsSkill
- domain-buildSkill
- evidence-adjudicatorSkill
- executing-plansSkill
- external-cli-auditSkill
- feature-developmentSkill
- field-rename-verificationSkill
- finishing-a-development-branchSkill
- git-branch-conventionsSkill
- gitlab-cliSkill
- hotfix-charterSkill
- implementation-trackerSkill
- incident-revert-reviewSkill
- infra-config-investigatorSkill
- investigation-stateSkill
- kernel-splitSkill
- llm-behavior-investigatorSkill
- llm-skill-reviewSkill
- merge-authorization-gateSkill
- micro-harsh-reviewSkill
- output-verificationSkill
- pre-commit-gateSkill
- progressive-code-review-gateSkill
- progressive-harsh-reviewSkill
- providing-code-reviewSkill
- pr-triage-gateSkill
- push-authorization-gateSkill
- receiving-code-reviewSkill
- reproduction-experiment-investigatorSkill
- requesting-code-reviewSkill
- requirements-validation-pmSkill
- requirements-validationSkill
- scope-tripwireSkill
- session-handoffSkill
- skills-hierarchy-tuningSkill
- sp-bughuntSkill
- state-consistency-investigatorSkill
- subagent-driven-developmentSkill
- systematic-debuggingSkill
- test-driven-developmentSkill
- timeline-trace-investigatorSkill
- token-estimationSkill
- unified-commit-gateSkill
- using-git-worktreesSkill
- using-superpowersSkill
- verification-before-completionSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
5 rules. 0 are common, 5 are distinctive, and 0 are very specific to this project.
0
Common
Most of the 19 edited versions we read keep these.
5
Distinctive
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 61
Use before committing to a non-trivial architecture, data model, integration, or other significant design choice.
- Distinctive: line 63
Do not use after implementation has begun, for bug fixing, or for initial idea exploration.
- Distinctive: line 78
Verify fixes landed: every claimed resolution must be visible in the updated artifact.
- Distinctive: line 79
Keep each review answer to one sentence and each later round delta-only.
- Distinctive: line 80
Never claim convergence after the initial review.
“Common” is measured: kept by at least 6 in 10 of the 19 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Reference index
Load only the sections required by the process. Preflight, comparison, and hostile review are mandatory; the remaining sections are on demand.
| Need | Read this reference section |
| Before every debate | Preflight routing |
| Mandatory before building the matrix | Comparison protocol |
| Before adversarial review | Hostile review protocol |
| When an output example is needed | Output and example |
| When choosing a neighboring workflow | Companion skills |
Design Triad
Wrong skill? Idea exploration ->
brainstorming. Requirements validation ->requirements-validation. Implementation planning ->plan-and-execute.
Announce at start: "I'm using the debate skill to evaluate design options."
When to use
Use before committing to a non-trivial architecture, data model, integration, or other significant design choice. Also use when the user asks for design alternatives or a comparison.
Do not use after implementation has begun, for bug fixing, or for initial idea exploration.
Mandatory process
- Preflight. Load
Preflight routingbefore every debate. Choose a route within 30 seconds; unresolved high-stakes inputs go to the user. - Generate. Produce a minimum three genuinely distinct, implementable options. Include status quo or a time-boxed spike as a real comparison anchor. If only one option emerges, invoke
think-twice. - Compare. Load
Comparison protocolbefore building the matrix. Use all six criteria, no more than five words per cell. Recommend in two or three sentences. A recommendation is not completion. - Review. Author and reviewer must be separate. Load
Hostile review protocol; use one separated reviewer for reversible work and at least two for irreversible or high-blast-radius work. Do NOT invokeprogressive-harsh-revieworquantitative-decision-gateas the reviewer because both can route back to debate. - Probe edge cases. List no more than ten failure modes, boundaries, tests, or defensive integration points surfaced by review.
- Iterate. Run
review -> fix -> verify the fix in the artifact -> re-review. Complete a minimum two full review rounds and no more than three. Exit only when a round finds no new material issue. If three rounds completed without convergence, escalate tothinking-orchestratorwith the unresolved findings rather than continuing to iterate. Do NOT continue beyond 3 rounds.
Hard gates
- Three options must be genuinely different, not superficial variations or two straw men around a preferred answer.
- A separated reviewer must test the chosen option and the integrity of the option set.
- Verify fixes landed: every claimed resolution must be visible in the updated artifact.
- Keep each review answer to one sentence and each later round delta-only.
- Never claim convergence after the initial review.
Failure Modes
| Failure | Fix |
| Only one option ever generated | Invoke think-twice; never proceed with fewer than three genuinely distinct options |
| Reviewer is the same pass/agent as the author | Dispatch a fresh sub-agent, or make an explicit role switch that discards the author's reasoning and starts fresh from the artifact text |
| Three rounds completed without convergence | Escalate to thinking-orchestrator with the unresolved findings — do not keep iterating past round 3 |
| A rejected option is a straw man, not one a competent engineer would propose | Replace it once and rerun comparison; the next review becomes round one |
| A claimed fix was never verified in the updated artifact | Re-read the artifact directly — a fix summary is not evidence the fix landed |
| Convergence claimed after only one review round | Never claim convergence after the initial review; a minimum of two full rounds is a hard gate, not a suggestion |
Rationalizations to reject
| Excuse | Required response |
| There is only one way | Invoke think-twice. |
| Other options are obviously wrong | Explain the loss in the matrix. |
| This is too simple | Test the assumptions anyway. |
| Review found nothing | Answer every hostile-review question. |
| There is no time | Compare now to avoid rework. |
| The recommendation is done | Complete review, edge cases, and iteration. |
| I reviewed my own design | Separate author and reviewer. |
| I documented the resolution | Verify it landed in the artifact. |
Output
Return a compact inline decision record containing: decision and rationale; options considered; rejected options and reasons; edge-case catalog; review findings and resolutions; open risks. Write under docs/superpowers/specs/ only when explicitly requested.
Reference loading
Load only the named section. Preflight routing is mandatory before Step 1, Comparison protocol before Step 3, and Hostile review protocol before Step 4. Load the output example only when needed. When an installed copy is found (any of .claude/.codex/.agents), loading requires the separately-provisioned ~/.codex/superpowers-plus checkout -- an install missing that shared dependency gets a loud section-loader missing failure, not silent wrong content.
<!-- kernel-split-reference-loader:start -->_ks_ref=""
_ks_loader=""
for _candidate in \
"$HOME/.claude/skills/sp-debate/reference.md" \
"$HOME/.codex/skills/sp-debate/reference.md" \
"$HOME/.agents/skills/sp-debate/reference.md"
do
if [ -r "$_candidate" ]; then _ks_ref="$_candidate"; break; fi
done
if [ -n "$_ks_ref" ]; then
_ks_loader="$HOME/.codex/superpowers-plus/tools/section-loader.sh"
else
_project_root="$(git rev-parse --show-toplevel 2>/dev/null || true)"
_source_dir="$_project_root/skills/engineering/debate"
if [ -n "$_project_root" ] && [ -r "$_source_dir/skill.md" ] && \
[ -r "$_source_dir/reference.md" ] && \
[ -r "$_project_root/tools/section-loader.sh" ]; then
_ks_ref="$_source_dir/reference.md"
_ks_loader="$_project_root/tools/section-loader.sh"
fi
fi
[ -r "$_ks_ref" ] || { printf 'reference missing\n' >&2; exit 1; }
[ -r "$_ks_loader" ] || { printf 'section-loader missing\n' >&2; exit 1; }
# Replace <section heading> below with one of the exact strings from
# the Reference index table above before running.
_section='<section heading>'
bash "$_ks_loader" "$_ks_ref" "$_section" \
|| { printf 'section not found: %s\n' "$_section" >&2; exit 1; }<!-- kernel-split-reference-loader:end -->Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

