mirror of
https://github.com/ruvnet/ruflo.git
synced 2026-09-14 14:01:28 +08:00
rUv
c983c0d809
From the deep-research map of NousResearch/Hermes-Agent vs ruflo (3 small, high-confidence, in-repo adoptions; the rest were SKIP/ADR-gated). - #8 Prompt caching: agent_execute's direct Anthropic call (agent-execute-core.ts — the path the cli actually uses) now marks the system prompt as an ephemeral cache breakpoint (~90% discount on cached input tokens, 5-min TTL). Also applied the system+trailing-message breakpoint strategy to @claude-flow/providers' anthropic-provider (separate package; gated by new config.promptCache, default on). - #14 Reasoning-tag scrub: scrubReasoningBlocks() strips <think>/<thinking>/ <reasoning>/<REASONING_SCRATCHPAD> blocks from trajectory action/result before DISTILL, so extended-thinking tokens don't contaminate pattern embeddings. Boundary-gated (leaves prose that merely mentions the tags untouched). - #6 Tool-loop circuit breaker: new tool-loop-guardrail.ts; post-command records (command, success), pre-command warns at 3 / blocks at 5 consecutive failures of the same command with a recovery hint. Orthogonal to the security tool-output-guardrail (injection). Advisory by default. Note: @claude-flow/providers is NOT consumed by the cli (separate published package), so the cli-effective #8 fix is in agent-execute-core.ts; the providers change ships on its own cadence. Verification: cli + providers build clean; 29/29 targeted tests pass (hermes-tier1 6 + bandit 8 + bug-cluster 15). Deep-research map also produced 5 corrections to ADR-113 (#1907).
71 lines
3.0 KiB
TypeScript
71 lines
3.0 KiB
TypeScript
/**
|
|
* Hermes-Agent Tier-1 adoptions (see docs/reviews/intelligence-system-audit-2026-05-29.md
|
|
* → Hermes capability map):
|
|
* #14 — reasoning-tag scrubbing keeps extended-thinking blocks out of the
|
|
* learning trajectory (so DISTILL doesn't embed reasoning tokens).
|
|
* #6 — tool-loop circuit breaker warns/blocks on repeated consecutive
|
|
* failures of the same command.
|
|
*/
|
|
|
|
import { describe, it, expect, beforeEach } from 'vitest';
|
|
import { scrubReasoningBlocks } from '../src/mcp-tools/hooks-tools.js';
|
|
import {
|
|
checkCommandLoop,
|
|
recordCommandOutcome,
|
|
_resetLoopHistory,
|
|
} from '../src/mcp-tools/tool-loop-guardrail.js';
|
|
|
|
describe('#14 — reasoning-tag scrubbing', () => {
|
|
it('strips paired thinking/reasoning blocks', () => {
|
|
expect(scrubReasoningBlocks('did it <think>secret CoT</think> ok')).toBe('did it ok');
|
|
expect(scrubReasoningBlocks('a<thinking>x</thinking>b')).toBe('ab');
|
|
expect(scrubReasoningBlocks('a<reasoning>y</reasoning>b')).toBe('ab');
|
|
expect(scrubReasoningBlocks('a<REASONING_SCRATCHPAD>z</REASONING_SCRATCHPAD>b')).toBe('ab');
|
|
});
|
|
it('leaves prose that merely mentions the tag untouched (boundary-gated)', () => {
|
|
expect(scrubReasoningBlocks('We document the <think> tag here')).toBe('We document the <think> tag here');
|
|
});
|
|
it('is a no-op on tag-free text and non-strings', () => {
|
|
expect(scrubReasoningBlocks('plain output')).toBe('plain output');
|
|
expect(scrubReasoningBlocks(undefined as unknown as string)).toBe(undefined);
|
|
});
|
|
});
|
|
|
|
describe('#6 — tool-loop circuit breaker', () => {
|
|
beforeEach(() => _resetLoopHistory());
|
|
|
|
it('allows until the warn threshold, then warns, then blocks', () => {
|
|
const cmd = 'npm run flaky';
|
|
expect(checkCommandLoop(cmd).verdict).toBe('allow');
|
|
recordCommandOutcome(cmd, false);
|
|
recordCommandOutcome(cmd, false);
|
|
expect(checkCommandLoop(cmd).verdict).toBe('allow'); // 2 fails
|
|
recordCommandOutcome(cmd, false);
|
|
expect(checkCommandLoop(cmd).verdict).toBe('warn'); // 3 fails
|
|
recordCommandOutcome(cmd, false);
|
|
recordCommandOutcome(cmd, false);
|
|
const v = checkCommandLoop(cmd);
|
|
expect(v.verdict).toBe('block'); // 5 fails
|
|
expect(v.consecutiveFailures).toBe(5);
|
|
expect(v.hint).toMatch(/failed 5/);
|
|
});
|
|
|
|
it('a success resets the streak', () => {
|
|
const cmd = 'pytest';
|
|
for (let i = 0; i < 5; i++) recordCommandOutcome(cmd, false);
|
|
expect(checkCommandLoop(cmd).verdict).toBe('block');
|
|
recordCommandOutcome(cmd, true);
|
|
expect(checkCommandLoop(cmd).verdict).toBe('allow');
|
|
});
|
|
|
|
it('interleaved other commands do not break the streak (exact-match only)', () => {
|
|
recordCommandOutcome('build', false);
|
|
recordCommandOutcome('ls', true); // different command
|
|
recordCommandOutcome('build', false);
|
|
recordCommandOutcome('git status', true);
|
|
recordCommandOutcome('build', false);
|
|
expect(checkCommandLoop('build').verdict).toBe('warn'); // 3 build failures, ls/git ignored
|
|
expect(checkCommandLoop('ls').verdict).toBe('allow');
|
|
});
|
|
});
|