Design session. Explore problem, scan codebase, generate 2-3 approaches, recommend one, produce .godmode/spec.md.
From godmodenpx claudepluginhub arbazkhan971/godmodeThis skill uses the workspace's default tool permissions.
Designs and optimizes AI agent action spaces, tool definitions, observation formats, error recovery, and context for higher task completion rates.
Enables AI agents to execute x402 payments with per-task budgets, spending controls, and non-custodial wallets via MCP tools. Use when agents pay for APIs, services, or other agents.
Compares coding agents like Claude Code and Aider on custom YAML-defined codebase tasks using git worktrees, measuring pass rate, cost, time, and consistency.
/godmode:think, "design", "brainstorm"PROBLEM STATEMENT:
- Goal: <one sentence>
- Constraints: <time, tech, backwards compat>
- Success criteria: <shell command that exits 0>
- Out of scope: <what this does NOT include>
IF ambiguous: offer 2 interpretations with examples
"A) You want X -- e.g. user clicks, modal appears"
"B) You want Y -- e.g. data syncs on page load"
User picks one. Proceed immediately.
IF success criteria is subjective ("works well"):
reject and propose concrete alternative
e.g., "response time < 200ms" or "tests pass"
IF greenfield (no repo): skip to Step 3.
# Project structure
git ls-files | head -150
# Entry points and configs
cat package.json 2>/dev/null \
|| cat pyproject.toml 2>/dev/null
# Existing types related to feature
grep -rn "interface\|type\|class\|def " \
--include="*.ts" --include="*.py" | head -40
# Recent changes in related areas
git log --oneline -15 -- <affected-paths>
# Test patterns
ls -la tests/ test/ __tests__/ spec/ 2>/dev/null
CODEBASE SCAN:
- Framework: <name + version>
- Patterns to reuse: <routing, error handling>
- Patterns to avoid: <deprecated patterns>
- Related code: <files in same domain>
APPROACH {A|B|C}: {name}
What: {1-2 sentences}
Why it wins: {main advantage}
Why it loses: {main disadvantage}
Files: {N} to create/modify
Disqualify any approach that:
RECOMMENDATION: Approach {A|B|C}
- Chosen because: <1-2 sentences>
- Approach B rejected: <key weakness>
- Top risk: <biggest risk>
- Mitigation: <concrete action>
IF two approaches are equal: prefer fewer files.
Write .godmode/spec.md with sections:
Problem (1-3 sentences), Approach (2-5),
Success Criteria (shell command exits 0),
Out of Scope, Files to Modify (path + what),
Files to Create (path + purpose),
Risks (risk + mitigation pairs).
Keep under 100 lines total.
# Verify referenced files exist
for f in <files_to_modify>; do
git ls-files --error-unmatch "$f"
done
# Verify new file parent dirs exist
for f in <files_to_create>; do
test -d "$(dirname "$f")" \
|| echo "MISSING: $(dirname "$f")"
done
# Verify success criteria syntax
bash -n -c '<success_criteria_cmd>' 2>&1
# Verify spec length (target <100)
wc -l .godmode/spec.md
# Commit
git add .godmode/spec.md \
&& git commit -m "spec: {feature}"
IF any validation fails: fix before committing.
Think: problem -- <goal>.
Think: scanned codebase -- <framework>, {N} files.
Think: generated {N} approaches.
Think: recommending Approach {X} -- <reason>.
Think: wrote .godmode/spec.md ({N} lines).
Spec: {path}. Approach: {chosen}. Files: {N}+{N}.
Append to .godmode/think-log.tsv:
timestamp\tfeature\tapproaches\tchosen\tfiles_modify\tfiles_create\tspec_lines\trisks
KEEP if: approach passes disqualification criteria
AND has testable success criteria
DISCARD if: violates constraints OR >30 files
OR no testable criteria
STOP when FIRST of:
- spec.md written, validated, and committed
- 3 approaches evaluated (max)
- >5 validation failures on spec