claude-code-synthesis

Generate a wide field of ideas, then pressure-test them. Five generators diverge, two challengers prune the weak ones, three synthesizers rank what survives.

Three-agent adversarial bug finder. Hunter overclaims every issue it can find, Skeptic disproves what it can, Referee arbitrates what survives. Scoring incentives force each role to behave honestly.

Dialectic process for reviewing, brainstorming, comparing tradeoffs, or stress-testing plans. Configurable modes, agent counts, and expert lenses.

Execute a pre-defined plan using implementation agents (no design changes)

Capture structured learnings (gotcha, pattern, decision, bug-fix) as JSONL per project. Cross-project searchable.

Assume the plan failed — then explain why. Three pessimists write failure narratives, two optimists rebut, a risk assessor weighs which failure modes are real.

Adversarial stress-test of code, a plan, or an argument. Three critics attack, one defender rebuts, a referee judges, then a hostile auditor attacks the synthesis itself.

Compare 2+ options with a dedicated advocate for each, counter-advocates who challenge every position, and a referee who ranks what survives.

Verify outputs after pipeline steps, implementations, or any multi-step work. Catches missing data, stale references, broken assumptions. Use after completing significant work, before declaring done.

Create or update all 4 COMP files (CLAUDE.md, ORIENT.md, MEMORY.md, PLAN.md) for a directory

Structured debugging with mandatory evidence gathering before any fix attempt. 4-phase protocol (Observe → Hypothesize → Reproduce → Fix) with hard gates between phases.

End-of-session COMP update — record decisions, progress, and state changes

Plan a new task/project using structured decomposition before any implementation

Launch a research-only agent that investigates a topic and returns findings (no implementation)

Periodic retrospective — assess what Claude did well, what went poorly, and what to improve

Review completed work against success criteria and flag issues

Pre-commit readiness check. Verifies tests, COMP freshness, uncommitted changes, then drafts commit message and stages specific files.

Start a Socratic debate session to stress-test a philosophical framework or thesis

End-of-session closer. Compiles session notes, gets fresh-context review, updates COMP files, checks skill health and execution gaps, runs bloat check.