Skip to content

Test.md

Latest commit

 

History

History
executable file
·
100 lines (77 loc) · 3.84 KB

Test.md

File metadata and controls

executable file
·
100 lines (77 loc) · 3.84 KB
Raw

Prompt to test H2C independently

Instructions for the AI:

Read the official H2C repository and run 5 tests completely independently.

Repository to consult:

Task:

Run 5 independent tests simulating H2C communication chains with increasing complexity.
Act as a receiving agent and respond exclusively in H2C format respecting the v1.4 specification.

The 5 tests to execute:

Test 1 – Simple Level
A simple "Hello World" project.
Must start with:

  • CTX:NEGOTIATE block (version handshake, REQUIRED first block)
  • STATE:ACK with protocol:h2c_v1.4

Test 2 – Medium Level
A CLI calculator with implementation, test, and a fix (cycle_id required in TEST:FAIL → BUILD:FIX → BUILD:DONE → TEST:PASS cycle).
Use v1.4 fields: cycle_id, retry_n:1, fail_count, pass_count, base_rev, rev.

Test 3 – Advanced Level
Refactoring towards Clean Architecture with context management (CTX:PRIMITIVES / CTX:UPDATE).
Use v1.4 features:

  • CTX:PRIMITIVES with ~task, ~constraint, ~goal, ~form
  • STATE:FINDINGS with formal fields: cause:, action:, impact:, components:

Test 4 – Very Complex Level
A long chain (minimum 10-12 messages) for a mini RAG pipeline (research, implementation, multiple tests, fixes, context management, and closure).
Exercise independent fix cycles with distinct cycle_id values.

Test 5 – Stress Test v1.4
Chain of 60+ messages to validate all v1.4 features. Verify:

  • CTX:NEGOTIATE emitted as first block — mandatory handshake
  • STATE:ACK with protocol:h2c_v1.4 immediately after
  • CTX:PRUNE emitted every 5 messages (follow the pruning table in SPEC.md §5.3)
  • CTX:COMPACT emitted every 20 messages
  • CTX:FREEZE emitted beyond msg 100 when COMPACT is no longer sufficient
  • BUILD:NACK emitted on at least one malformed block (with ref_id, error, hint)
  • STATE:FINDINGS uses formal fields: cause:, action:, impact:, risk:, components:, pattern:
  • DAG after: transitive closure validated (no cycles across N nodes)
  • v1.4 fields used correctly: rev:, after:, notes:, base_rev:, fail_count:, pass_count:, cycle_id:, retry_n:
  • Expected break point beyond msg 100 (model limit, not protocol limit)

Required output:

After executing the tests, produce a complete report with:

  1. Brief description of each test

  2. Summary table with columns:

    • Test
    • Complexity
    • Estimated Token Savings (%)
    • Comprehension
    • Stability
    • Break point (if reached)

  3. Main observations (strengths and weaknesses observed, especially on long chains):

    • How does CTX:NEGOTIATE affect multi-agent compatibility?
    • Is BUILD:NACK effective for error recovery?
    • Are formal STATE:FINDINGS fields parseable by the receiver?

  4. v1.3 vs v1.4 comparison with data on:

    • Version negotiation (absent vs CTX:NEGOTIATE)
    • Error recovery (silent discard vs BUILD:NACK)
    • FINDINGS structure (free-form vs formal fields)
    • DAG cycle detection (2-node vs transitive closure)
    • Break point and stability

  5. Conclusion on the behavior of H2C v1.4.

Important rules:

  • Really run the tests independently (don't simulate results, generate real chains).
  • Use only the H2C v1.4 format for internal test responses.
  • Every chain must start with CTX:NEGOTIATE followed by STATE:ACK.
  • On malformed blocks, emit BUILD:NACK instead of silent discard.
  • For Test 5, apply PRUNE every 5 messages and COMPACT every 20 messages, FREEZE beyond msg 100.
  • STATE:FINDINGS should use formal fields: cause:, action:, impact:.
  • Be objective and technical in the results.
  • At the end of the report indicate the test date.

Start when ready.