Skip to content

moonlight-lupin/agent-skills / plugins/skill-retrieval

skill-retrieval

Verified

BM25-based skill retrieval plugin for Hermes Agent. Replaces the full skill list in the system prompt with a names-only compact view (~2K tokens) and injects top-K relevant skill descriptions per turn via BM25 retrieval (~300 tokens). Saves ~9K tokens/turn. Use when system prompt token overhead from skills is a concern, or when skill discovery quality matters.

tests/test_jev_environ.py

View on GitHub
# ═══════════════════════════════════════════════════════════════════════════════
# Environ-only resolution (code-review finding #1 on PR 131290)
# ═══════════════════════════════════════════════════════════════════════════════

import importlib
import sys
from pathlib import Path

SCRIPTS = Path(__file__).resolve().parent.parent / "scripts"


def _fresh_mod():
    """Import (or re-import) the reranker module from scripts/."""
    if str(SCRIPTS) not in sys.path:
        sys.path.insert(0, str(SCRIPTS))
    mod = importlib.import_module("jev_rerank")
    return importlib.reload(mod)


def test_load_key_reads_environ_only(monkeypatch, tmp_path):
    """_load_key must read os.environ only.

    Hazard: reading the Hermes .env file directly fails the security scan and,
    under a non-default profile whose startup env differs, can read another
    profile's secrets. Hermes loads the active profile's .env into os.environ
    at startup, so environ is the single correct source.
    """
    import os
    env_file = tmp_path / ".env"
    env_file.write_text("TYPESAFE_API_KEY=from-file\nTYPESAFE_KEY=legacy-from-file\n")
    monkeypatch.setenv("HERMES_HOME", str(tmp_path))
    monkeypatch.setenv("HOME", str(tmp_path))
    monkeypatch.delenv("TYPESAFE_API_KEY", raising=False)
    monkeypatch.delenv("TYPESAFE_KEY", raising=False)
    mod = _fresh_mod()
    assert mod._load_key() is None  # file must NOT be consulted


def test_load_key_environ_primary(monkeypatch, tmp_path):
    """TYPESAFE_API_KEY wins over legacy TYPESAFE_KEY; None when absent."""
    monkeypatch.setenv("HERMES_HOME", str(tmp_path))
    monkeypatch.setenv("HOME", str(tmp_path))
    monkeypatch.delenv("TYPESAFE_API_KEY", raising=False)
    monkeypatch.setenv("TYPESAFE_KEY", "legacy-k")
    mod = _fresh_mod()
    assert mod._load_key() == "legacy-k"
    monkeypatch.setenv("TYPESAFE_API_KEY", "primary-k")
    assert mod._load_key() == "primary-k"


def test_rerank_env_flag_environ_only(monkeypatch, tmp_path):
    """SKILL_RETRIEVAL_RERANK is read from environ only, never from a .env file."""
    env_file = tmp_path / ".env"
    env_file.write_text("SKILL_RETRIEVAL_RERANK=jev\n")
    monkeypatch.setenv("HERMES_HOME", str(tmp_path))
    monkeypatch.setenv("HOME", str(tmp_path))
    monkeypatch.delenv("SKILL_RETRIEVAL_RERANK", raising=False)
    mod = _fresh_mod()
    assert mod.RERANK_MODE == "off"  # file must NOT enable rerank


def test_env_log_path_uses_hermes_home_helper(monkeypatch, tmp_path):
    """Default A/B log path resolves through hermes_constants.get_hermes_home().

    Hazard: a hardcoded Path.home()/.hermes path ignores profile homes and the
    HERMES_HOME override, so logs for different profiles can collide or land
    outside the profile's data dir.
    """
    monkeypatch.setenv("HERMES_HOME", str(tmp_path))
    monkeypatch.delenv("SKILL_RETRIEVAL_RERANK_LOG", raising=False)
    mod = _fresh_mod()
    p = mod._env_log_path()
    assert str(p).endswith("data/jev-trial/rerank_ab_log.jsonl") or str(
        p
    ).endswith("data\\jev-trial\\rerank_ab_log.jsonl")
    assert str(tmp_path) in str(p)


def test_ab_log_rotates_when_over_cap(monkeypatch, tmp_path):
    """A/B log past LOG_MAX_BYTES rotates to <name>.1 (one older generation).

    Hazard: an unattended install could accumulate an unbounded JSONL file.
    """
    log_path = tmp_path / "rerank_ab_log.jsonl"
    monkeypatch.setenv("SKILL_RETRIEVAL_RERANK", "jev")
    monkeypatch.setenv("SKILL_RETRIEVAL_RERANK_LOG", str(log_path))
    mod = _fresh_mod()
    mod.LOG_MAX_BYTES = 100  # shrink the cap for the test
    old = {"ts": "t0", "query": "x" * 40}
    new = {"ts": "t1", "query": "y" * 40}
    mod._write_log(old)
    mod._write_log(new)  # should push past the 100-byte cap
    mod._write_log(new)  # triggers rotation
    rotated = tmp_path / "rerank_ab_log.jsonl.1"
    assert log_path.exists()
    assert rotated.exists(), "log must rotate once over the cap"
    assert "t0" in rotated.read_text()  # pre-rotation content survived
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall