mcpField
D
dialectic

dialectic

github.com/umputunAgent workflows49061 installsUpdated 3d ago
Repository

About dialectic

dialectic is an analysis skill for evaluating claims through two agents working in parallel: one examines supporting evidence, the other opposing evidence. It then synthesizes and verifies their findings against the code.

Key features

  1. 01Parallel thesis and antithesis analysis
  2. 02Evidence with file paths and line numbers
  3. 03Synthesis of opposing findings
  4. 04Verification against implementation
  5. 05Supports architecture, bug, and code reviews

Use cases

  1. 01Assessing architecture decisions
  2. 02Evaluating bug-fix claims
  3. 03Checking performance assertions
  4. 04Reviewing code changes

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/umputun/cc-thingz --skill dialectic

After installing, type/dialecticin Claude Code to confirm it loaded.

Details

From the project README

Dialectic Analysis

Objective analysis of a statement by running two agents with opposing goals in parallel, then synthesizing findings.

Process

Step 1: Launch Parallel Agents

CRITICAL: Both Task tool calls MUST be in a single message for true parallel execution. Do NOT use run_in_background. Do NOT launch sequentially. Foreground agents in the same message run in parallel and block until both complete.

Agent 1 (Thesis) — find all POSITIVE evidence:

  • what works well, improvements, correct patterns
  • supporting facts, proof the statement is TRUE
  • benefits, strengths, successful outcomes

Agent 2 (Antithesis) — find all NEGATIVE evidence:

  • what's problematic, issues, risks, anti-patterns
  • edge cases, proof the statement is FALSE
  • weaknesses, failure modes, hidden costs

Both agents must provide specific file paths and line numbers when analyzing code.

Step 2: Synthesize

After both agents complete, synthesize findings into an objective conclusion:

  • weigh evidence from both sides
  • identify where thesis and antithesis agree (strongest signal)
  • note unresolved tensions
Step 3: Verify

CRITICAL — after presenting the synthesis, verify it against actual implementation:

  • read the specific files and line numbers referenced by both agents
  • confirm the evidence cited actually exists in the codebase
  • check the implementation flow matches the claims made
  • verify the synthesis accurately reflects the actual code context
  • identify any misinterpretations or missing context from agent reports
  • revise synthesis if verification reveals inaccuracies

Use Cases

Architecture decisions:

/thinking-tools:dialectic this microservice split improves maintainability

Bug analysis:

/thinking-tools:dialectic the connection pool fixes the timeout issue

Performance claims:

/thinking-tools:dialectic caching reduced database load by 80%

Refactoring safety:

/thinking-tools:dialectic extracting this interface simplifies testing

Code review:

/thinking-tools:dialectic this implementation is thread-safe

Review changes:

/thinking-tools:dialectic review the changes in server.go

Key Principles

  • Eliminate confirmation bias — examining both sides simultaneously prevents anchoring on the first conclusion
  • Evidence-based — agents must cite specific files, lines, and facts, not general claims
  • Verification required — synthesis must be checked against actual code before presenting
  • Objective conclusion — the goal is truth, not winning either side

Comments

Markdown supported