Compare commits

..

2 Commits

9 changed files with 591 additions and 914 deletions

View File

@@ -1,114 +0,0 @@
# Knowledge File Format Specification
**Version:** 1
**Issue:** #10
**Status:** Draft
---
## Overview
The knowledge system has two layers:
1. **index.json** — Machine-readable fact index. Fast lookups by ID, category, repo, tags.
2. **Knowledge files** (YAML) — Human-readable, editable facts organized by domain.
The harvester writes to both. The bootstrapper reads from index.json. Humans edit the YAML files directly.
---
## index.json Schema
```json
{
"version": 1,
"last_updated": "ISO-8601 timestamp",
"total_facts": 0,
"facts": []
}
```
### Fact Object
| Field | Type | Required | Description |
|-------|------|----------|-------------|
| `id` | string | yes | Unique identifier: `{domain}:{category}:{sequence}` |
| `fact` | string | yes | One-sentence description of the knowledge |
| `category` | enum | yes | One of: `fact`, `pitfall`, `pattern`, `tool-quirk`, `question` |
| `domain` | string | yes | Where this applies: repo name, `global`, or agent name |
| `confidence` | float | yes | 0.01.0. How certain is this knowledge? |
| `tags` | string[] | no | Searchable labels |
| `source_count` | int | no | How many sessions confirmed this fact |
| `first_seen` | date | no | ISO-8601 date first extracted |
| `last_confirmed` | date | no | ISO-8601 date last seen in a session |
| `expires` | date | no | Optional. After this date, fact is stale |
| `related` | string[] | no | IDs of related facts |
### ID Format: `{domain}:{category}:{sequence}`
### Categories
| Category | Definition |
|----------|------------|
| `fact` | Concrete, verifiable information |
| `pitfall` | Errors, wrong assumptions, time-wasters |
| `pattern` | Successful sequences of actions |
| `tool-quirk` | Environment-specific behaviors |
| `question` | Identified but unanswered |
### Confidence Scoring
| Range | Meaning |
|-------|---------|
| 0.91.0 | Explicitly stated and verified |
| 0.70.8 | Clearly implied by multiple data points |
| 0.50.6 | Suggested but not fully verified |
| 0.30.4 | Inferred from limited data |
| 0.10.2 | Speculative or uncertain |
---
## Directory Structure
```
knowledge/
├── index.json # Machine-readable fact index
├── SCHEMA.md # This file
├── global/ # Cross-repo knowledge
│ ├── pitfalls.yaml
│ ├── patterns.yaml
│ └── tool-quirks.yaml
├── repos/ # Per-repo knowledge
│ ├── {repo-name}.yaml
│ └── ...
└── agents/ # Agent-type knowledge
└── {agent-type}.yaml
```
## YAML File Format
YAML files use frontmatter for metadata, then markdown sections with fact entries:
```yaml
---
domain: global
category: tool-quirk
version: 1
last_updated: "2026-04-13"
---
# Title
## Section
- id: global:tool-quirk:001
fact: "Description"
confidence: 0.95
tags: [tag1, tag2]
source_count: 5
first_seen: "2026-03-27"
```
## Validation
Run `python scripts/validate_knowledge.py` to validate index.json.

View File

@@ -1,80 +0,0 @@
---
domain: global
category: pitfall
version: 1
last_updated: "2026-04-13"
---
# Pitfalls (Global)
Cross-repo traps that waste time across the fleet.
## Git & Forge
- id: global:pitfall:001
fact: "Branch protection requires 1 approval on main - API merges fail with 405 without it"
confidence: 0.95
tags: [git, merge, branch-protection, gitea]
source_count: 12
first_seen: "2026-04-05"
last_confirmed: "2026-04-13"
related: [the-nexus:pitfall:001]
- id: global:pitfall:002
fact: "Never use --no-verify on git commits - it bypasses all hooks including safety checks"
confidence: 0.95
tags: [git, hooks, safety]
source_count: 5
first_seen: "2026-03-28"
last_confirmed: "2026-04-13"
- id: global:pitfall:003
fact: "Gitea PR creation workaround needed on the-nexus - direct API call fails, use alternative endpoint"
confidence: 0.9
tags: [gitea, pr, api, workaround]
source_count: 4
first_seen: "2026-04-06"
last_confirmed: "2026-04-12"
## Agent Operations
- id: global:pitfall:004
fact: "Anthropic is BANNED from fallback chain - if fallback triggers to Anthropic, something is wrong"
confidence: 0.95
tags: [provider, anthropic, fallback]
source_count: 7
first_seen: "2026-03-30"
last_confirmed: "2026-04-13"
- id: global:pitfall:005
fact: "Telegram tokens expired - don't assume Telegram notifications work without checking"
confidence: 0.85
tags: [telegram, notifications, token]
source_count: 3
first_seen: "2026-04-02"
- id: global:pitfall:006
fact: "Multiple gateways = 'cannot schedule futures' error - only one gateway process should run"
confidence: 0.9
tags: [gateway, cron, process]
source_count: 4
first_seen: "2026-04-04"
last_confirmed: "2026-04-11"
## Testing
- id: global:pitfall:007
fact: "pytest root collection picks up operational *_test.py scripts - restrict to tests/ directory"
confidence: 0.9
tags: [pytest, test, collection]
source_count: 3
first_seen: "2026-04-07"
last_confirmed: "2026-04-13"
- id: global:pitfall:008
fact: "TDD: test 1 before building 55 - verify the cycle works before scaling"
confidence: 0.95
tags: [tdd, testing, methodology]
source_count: 8
first_seen: "2026-03-25"
last_confirmed: "2026-04-13"

View File

@@ -1,71 +0,0 @@
---
domain: global
category: tool-quirk
version: 1
last_updated: "2026-04-13"
---
# Tool Quirks (Global)
## Authentication
- id: global:tool-quirk:001
fact: "Gitea token stored at ~/.config/gitea/token, not env var GITEA_TOKEN"
confidence: 0.95
tags: [git, auth, gitea, token]
source_count: 23
first_seen: "2026-03-27"
last_confirmed: "2026-04-13"
related: [global:pitfall:001]
- id: global:tool-quirk:002
fact: "Gitea API uses 'Authorization: token TOKEN' header format, not Bearer"
confidence: 0.9
tags: [git, api, gitea]
source_count: 8
first_seen: "2026-03-28"
last_confirmed: "2026-04-12"
- id: global:tool-quirk:003
fact: "Gitea Issues API type=issues param does NOT filter PRs - use truthiness check on pull_request field"
confidence: 0.95
tags: [gitea, api, issues, pr]
source_count: 6
first_seen: "2026-04-01"
last_confirmed: "2026-04-13"
## Paths & Environment
- id: global:tool-quirk:004
fact: "~/.hermes is the default hermes home - check get_hermes_home() not the path literal"
confidence: 0.9
tags: [paths, hermes, env]
source_count: 10
first_seen: "2026-03-30"
last_confirmed: "2026-04-13"
related: [hermes-agent:pitfall:005]
- id: global:tool-quirk:005
fact: "Ansible vault-encrypted vars in YAML require vault_inline_vars plugin"
confidence: 0.85
tags: [ansible, vault, config]
source_count: 3
first_seen: "2026-04-02"
## Model & Inference
- id: global:tool-quirk:006
fact: "mimo-v2-pro via Nous Research is the default model - don't assume Anthropic is available"
confidence: 0.95
tags: [model, provider, nous, default]
source_count: 15
first_seen: "2026-03-25"
last_confirmed: "2026-04-13"
- id: global:tool-quirk:007
fact: "Kill + restart with 'hermes chat' preserves old model state - NEVER use --resume"
confidence: 0.95
tags: [hermes, model, restart, session]
source_count: 8
first_seen: "2026-03-29"
last_confirmed: "2026-04-12"

View File

@@ -1,472 +1,6 @@
{
"version": 1,
"last_updated": "2026-04-13T20:00:00Z",
"total_facts": 29,
"facts": [
{
"id": "hermes-agent:pitfall:001",
"fact": "deploy-crons.py leaves jobs in mixed model format",
"category": "pitfall",
"domain": "hermes-agent",
"confidence": 0.95,
"tags": [
"cron",
"deploy",
"model",
"config"
],
"source_count": 5,
"first_seen": "2026-04-08",
"last_confirmed": "2026-04-13",
"related": [
"hermes-agent:pitfall:002",
"hermes-agent:pitfall:003"
]
},
{
"id": "hermes-agent:pitfall:002",
"fact": "deploy-crons.py --deploy doesn't set legacy skill field from skills list",
"category": "pitfall",
"domain": "hermes-agent",
"confidence": 0.9,
"tags": [
"cron",
"deploy",
"skills"
],
"source_count": 3,
"first_seen": "2026-04-09",
"last_confirmed": "2026-04-13",
"related": [
"hermes-agent:pitfall:001"
]
},
{
"id": "hermes-agent:pitfall:003",
"fact": "Cron jobs with blank fallback_model fields trigger spurious gateway warnings",
"category": "pitfall",
"domain": "hermes-agent",
"confidence": 0.9,
"tags": [
"cron",
"model",
"fallback"
],
"source_count": 4,
"first_seen": "2026-04-07",
"last_confirmed": "2026-04-12",
"related": [
"hermes-agent:pitfall:001"
]
},
{
"id": "hermes-agent:pitfall:004",
"fact": "model-watchdog.py checks first provider line, not model.provider - causes false drift alarms",
"category": "pitfall",
"domain": "hermes-agent",
"confidence": 0.9,
"tags": [
"watchdog",
"model",
"config"
],
"source_count": 3,
"first_seen": "2026-04-08",
"last_confirmed": "2026-04-13"
},
{
"id": "hermes-agent:pitfall:005",
"fact": "10+ files read HERMES_HOME directly instead of get_hermes_home()",
"category": "pitfall",
"domain": "hermes-agent",
"confidence": 0.85,
"tags": [
"paths",
"env",
"hermes-home"
],
"source_count": 6,
"first_seen": "2026-04-06",
"last_confirmed": "2026-04-12",
"related": [
"global:pitfall:002"
]
},
{
"id": "hermes-agent:pitfall:006",
"fact": "get_hermes_home() doesn't expand tilde when HERMES_HOME=~/... is set",
"category": "pitfall",
"domain": "hermes-agent",
"confidence": 0.8,
"tags": [
"paths",
"env",
"bug"
],
"source_count": 2,
"first_seen": "2026-04-05"
},
{
"id": "hermes-agent:pitfall:007",
"fact": "vps-agent-dispatch reports OK while remote hermes binary path is broken",
"category": "pitfall",
"domain": "hermes-agent",
"confidence": 0.9,
"tags": [
"ssh",
"dispatch",
"vps"
],
"source_count": 4,
"first_seen": "2026-04-07",
"last_confirmed": "2026-04-11"
},
{
"id": "hermes-agent:pitfall:008",
"fact": "nightwatch-health-monitor SSH check fails on cloud-model-only deployments",
"category": "pitfall",
"domain": "hermes-agent",
"confidence": 0.85,
"tags": [
"ssh",
"health",
"cloud"
],
"source_count": 2,
"first_seen": "2026-04-10"
},
{
"id": "the-nexus:pitfall:001",
"fact": "Merges fail with HTTP 405 due to branch protection",
"category": "pitfall",
"domain": "the-nexus",
"confidence": 0.95,
"tags": [
"git",
"merge",
"branch-protection",
"gitea"
],
"source_count": 12,
"first_seen": "2026-04-05",
"last_confirmed": "2026-04-13",
"related": [
"global:pitfall:001"
]
},
{
"id": "the-nexus:pitfall:002",
"fact": "ThreadingHTTPServer required for multi-user bridge - standard HTTPServer blocks on concurrent requests",
"category": "pitfall",
"domain": "the-nexus",
"confidence": 0.95,
"tags": [
"server",
"concurrency",
"bridge"
],
"source_count": 5,
"first_seen": "2026-04-10",
"last_confirmed": "2026-04-13"
},
{
"id": "the-nexus:pitfall:003",
"fact": "ChatLog.log() crashes on message persistence when index.html has orphaned button tags",
"category": "pitfall",
"domain": "the-nexus",
"confidence": 0.9,
"tags": [
"html",
"crash",
"chatlog"
],
"source_count": 3,
"first_seen": "2026-04-12",
"last_confirmed": "2026-04-13"
},
{
"id": "the-nexus:pitfall:004",
"fact": "Three.js LOD not implemented - local hardware struggles with full scene",
"category": "pitfall",
"domain": "the-nexus",
"confidence": 0.85,
"tags": [
"threejs",
"performance",
"lod"
],
"source_count": 4,
"first_seen": "2026-04-09",
"last_confirmed": "2026-04-13"
},
{
"id": "the-nexus:pitfall:005",
"fact": "Duplicate content blocks appear in index.html when PR merges conflict silently",
"category": "pitfall",
"domain": "the-nexus",
"confidence": 0.8,
"tags": [
"html",
"merge-conflict",
"duplicate"
],
"source_count": 3,
"first_seen": "2026-04-11",
"last_confirmed": "2026-04-13"
},
{
"id": "the-nexus:pitfall:006",
"fact": "Unified HTTP + WebSocket server required for proper URL deployment - separate servers break CORS",
"category": "pitfall",
"domain": "the-nexus",
"confidence": 0.9,
"tags": [
"deploy",
"websocket",
"http",
"cors"
],
"source_count": 4,
"first_seen": "2026-04-10",
"last_confirmed": "2026-04-13"
},
{
"id": "global:tool-quirk:001",
"fact": "Gitea token stored at ~/.config/gitea/token, not env var GITEA_TOKEN",
"category": "tool-quirk",
"domain": "global",
"confidence": 0.95,
"tags": [
"git",
"auth",
"gitea",
"token"
],
"source_count": 23,
"first_seen": "2026-03-27",
"last_confirmed": "2026-04-13",
"related": [
"global:pitfall:001"
]
},
{
"id": "global:tool-quirk:002",
"fact": "Gitea API uses 'Authorization: token TOKEN' header format, not Bearer",
"category": "tool-quirk",
"domain": "global",
"confidence": 0.9,
"tags": [
"git",
"api",
"gitea"
],
"source_count": 8,
"first_seen": "2026-03-28",
"last_confirmed": "2026-04-12"
},
{
"id": "global:tool-quirk:003",
"fact": "Gitea Issues API type=issues param does NOT filter PRs",
"category": "tool-quirk",
"domain": "global",
"confidence": 0.95,
"tags": [
"gitea",
"api",
"issues",
"pr"
],
"source_count": 6,
"first_seen": "2026-04-01",
"last_confirmed": "2026-04-13"
},
{
"id": "global:tool-quirk:004",
"fact": "~/.hermes is the default hermes home - check get_hermes_home() not the path literal",
"category": "tool-quirk",
"domain": "global",
"confidence": 0.9,
"tags": [
"paths",
"hermes",
"env"
],
"source_count": 10,
"first_seen": "2026-03-30",
"last_confirmed": "2026-04-13",
"related": [
"hermes-agent:pitfall:005"
]
},
{
"id": "global:tool-quirk:005",
"fact": "Ansible vault-encrypted vars in YAML require vault_inline_vars plugin",
"category": "tool-quirk",
"domain": "global",
"confidence": 0.85,
"tags": [
"ansible",
"vault",
"config"
],
"source_count": 3,
"first_seen": "2026-04-02"
},
{
"id": "global:tool-quirk:006",
"fact": "mimo-v2-pro via Nous Research is the default model - don't assume Anthropic is available",
"category": "tool-quirk",
"domain": "global",
"confidence": 0.95,
"tags": [
"model",
"provider",
"nous",
"default"
],
"source_count": 15,
"first_seen": "2026-03-25",
"last_confirmed": "2026-04-13"
},
{
"id": "global:tool-quirk:007",
"fact": "Kill + restart with 'hermes chat' preserves old model state - NEVER use --resume",
"category": "tool-quirk",
"domain": "global",
"confidence": 0.95,
"tags": [
"hermes",
"model",
"restart",
"session"
],
"source_count": 8,
"first_seen": "2026-03-29",
"last_confirmed": "2026-04-12"
},
{
"id": "global:pitfall:001",
"fact": "Branch protection requires 1 approval on main - API merges fail with 405 without it",
"category": "pitfall",
"domain": "global",
"confidence": 0.95,
"tags": [
"git",
"merge",
"branch-protection",
"gitea"
],
"source_count": 12,
"first_seen": "2026-04-05",
"last_confirmed": "2026-04-13",
"related": [
"the-nexus:pitfall:001"
]
},
{
"id": "global:pitfall:002",
"fact": "Never use --no-verify on git commits",
"category": "pitfall",
"domain": "global",
"confidence": 0.95,
"tags": [
"git",
"hooks",
"safety"
],
"source_count": 5,
"first_seen": "2026-03-28",
"last_confirmed": "2026-04-13"
},
{
"id": "global:pitfall:003",
"fact": "Gitea PR creation workaround needed on the-nexus - direct API call fails",
"category": "pitfall",
"domain": "global",
"confidence": 0.9,
"tags": [
"gitea",
"pr",
"api",
"workaround"
],
"source_count": 4,
"first_seen": "2026-04-06",
"last_confirmed": "2026-04-12"
},
{
"id": "global:pitfall:004",
"fact": "Anthropic is BANNED from fallback chain",
"category": "pitfall",
"domain": "global",
"confidence": 0.95,
"tags": [
"provider",
"anthropic",
"fallback"
],
"source_count": 7,
"first_seen": "2026-03-30",
"last_confirmed": "2026-04-13"
},
{
"id": "global:pitfall:005",
"fact": "Telegram tokens expired - don't assume Telegram notifications work",
"category": "pitfall",
"domain": "global",
"confidence": 0.85,
"tags": [
"telegram",
"notifications",
"token"
],
"source_count": 3,
"first_seen": "2026-04-02"
},
{
"id": "global:pitfall:006",
"fact": "Multiple gateways = 'cannot schedule futures' error - only one gateway process should run",
"category": "pitfall",
"domain": "global",
"confidence": 0.9,
"tags": [
"gateway",
"cron",
"process"
],
"source_count": 4,
"first_seen": "2026-04-04",
"last_confirmed": "2026-04-11"
},
{
"id": "global:pitfall:007",
"fact": "pytest root collection picks up operational *_test.py scripts - restrict to tests/ directory",
"category": "pitfall",
"domain": "global",
"confidence": 0.9,
"tags": [
"pytest",
"test",
"collection"
],
"source_count": 3,
"first_seen": "2026-04-07",
"last_confirmed": "2026-04-13"
},
{
"id": "global:pitfall:008",
"fact": "TDD: test 1 before building 55",
"category": "pitfall",
"domain": "global",
"confidence": 0.95,
"tags": [
"tdd",
"testing",
"methodology"
],
"source_count": 8,
"first_seen": "2026-03-25",
"last_confirmed": "2026-04-13"
}
]
"total_facts": 0,
"facts": []
}

View File

@@ -1,80 +0,0 @@
---
domain: hermes-agent
category: pitfall
version: 1
last_updated: "2026-04-13"
---
# Pitfalls (hermes-agent)
## Cron & Deployment
- id: hermes-agent:pitfall:001
fact: "deploy-crons.py leaves jobs in mixed model format - some have provider/model, some just model"
confidence: 0.95
tags: [cron, deploy, model, config]
source_count: 5
first_seen: "2026-04-08"
last_confirmed: "2026-04-13"
related: [hermes-agent:pitfall:002, hermes-agent:pitfall:003]
- id: hermes-agent:pitfall:002
fact: "deploy-crons.py --deploy doesn't set legacy skill field from skills list"
confidence: 0.9
tags: [cron, deploy, skills]
source_count: 3
first_seen: "2026-04-09"
last_confirmed: "2026-04-13"
related: [hermes-agent:pitfall:001]
- id: hermes-agent:pitfall:003
fact: "Cron jobs with blank fallback_model fields trigger spurious gateway warnings"
confidence: 0.9
tags: [cron, model, fallback]
source_count: 4
first_seen: "2026-04-07"
last_confirmed: "2026-04-12"
related: [hermes-agent:pitfall:001]
- id: hermes-agent:pitfall:004
fact: "model-watchdog.py checks first provider line, not model.provider - causes false drift alarms"
confidence: 0.9
tags: [watchdog, model, config]
source_count: 3
first_seen: "2026-04-08"
last_confirmed: "2026-04-13"
## Path & Environment
- id: hermes-agent:pitfall:005
fact: "10+ files read HERMES_HOME directly instead of get_hermes_home() - breaks on custom paths"
confidence: 0.85
tags: [paths, env, hermes-home]
source_count: 6
first_seen: "2026-04-06"
last_confirmed: "2026-04-12"
related: [global:pitfall:002]
- id: hermes-agent:pitfall:006
fact: "get_hermes_home() doesn't expand tilde when HERMES_HOME=~/... is set"
confidence: 0.8
tags: [paths, env, bug]
source_count: 2
first_seen: "2026-04-05"
## SSH & Dispatch
- id: hermes-agent:pitfall:007
fact: "vps-agent-dispatch reports OK while remote hermes binary path is broken"
confidence: 0.9
tags: [ssh, dispatch, vps]
source_count: 4
first_seen: "2026-04-07"
last_confirmed: "2026-04-11"
- id: hermes-agent:pitfall:008
fact: "nightwatch-health-monitor SSH check fails on cloud-model-only deployments"
confidence: 0.85
tags: [ssh, health, cloud]
source_count: 2
first_seen: "2026-04-10"

View File

@@ -1,63 +0,0 @@
---
domain: the-nexus
category: pitfall
version: 1
last_updated: "2026-04-13"
---
# Pitfalls (the-nexus)
## Git & Merging
- id: the-nexus:pitfall:001
fact: "Merges fail with HTTP 405 due to branch protection - must use merge API with 1 approval"
confidence: 0.95
tags: [git, merge, branch-protection, gitea]
source_count: 12
first_seen: "2026-04-05"
last_confirmed: "2026-04-13"
related: [global:pitfall:001]
- id: the-nexus:pitfall:002
fact: "ThreadingHTTPServer required for multi-user bridge - standard HTTPServer blocks on concurrent requests"
confidence: 0.95
tags: [server, concurrency, bridge]
source_count: 5
first_seen: "2026-04-10"
last_confirmed: "2026-04-13"
- id: the-nexus:pitfall:003
fact: "ChatLog.log() crashes on message persistence when index.html has orphaned button tags"
confidence: 0.9
tags: [html, crash, chatlog]
source_count: 3
first_seen: "2026-04-12"
last_confirmed: "2026-04-13"
## Three.js & Performance
- id: the-nexus:pitfall:004
fact: "Three.js LOD not implemented - local hardware struggles with full scene without texture optimization"
confidence: 0.85
tags: [threejs, performance, lod]
source_count: 4
first_seen: "2026-04-09"
last_confirmed: "2026-04-13"
- id: the-nexus:pitfall:005
fact: "Duplicate content blocks appear in index.html when PR merges conflict silently"
confidence: 0.8
tags: [html, merge-conflict, duplicate]
source_count: 3
first_seen: "2026-04-11"
last_confirmed: "2026-04-13"
## Deployment
- id: the-nexus:pitfall:006
fact: "Unified HTTP + WebSocket server required for proper URL deployment - separate servers break CORS"
confidence: 0.9
tags: [deploy, websocket, http, cors]
source_count: 4
first_seen: "2026-04-10"
last_confirmed: "2026-04-13"

447
scripts/harvester.py Normal file
View File

@@ -0,0 +1,447 @@
#!/usr/bin/env python3
"""
harvester.py — Extract durable knowledge from Hermes session transcripts.
Combines session_reader + extraction prompt + LLM inference to pull
facts, pitfalls, patterns, and tool quirks from finished sessions.
Usage:
python3 harvester.py --session ~/.hermes/sessions/session_xxx.jsonl --output knowledge/
python3 harvester.py --batch --since 2026-04-01 --limit 100
python3 harvester.py --session session.jsonl --dry-run # Preview without writing
"""
import argparse
import json
import os
import sys
import time
import hashlib
from datetime import datetime, timezone
from pathlib import Path
from typing import Optional
# Add scripts dir to path for sibling imports
SCRIPT_DIR = Path(__file__).parent.absolute()
sys.path.insert(0, str(SCRIPT_DIR))
from session_reader import read_session, extract_conversation, truncate_for_context, messages_to_text
# --- Configuration ---
DEFAULT_API_BASE = os.environ.get("HARVESTER_API_BASE", "https://api.nousresearch.com/v1")
DEFAULT_API_KEY = os.environ.get("HARVESTER_API_KEY", "")
DEFAULT_MODEL = os.environ.get("HARVESTER_MODEL", "xiaomi/mimo-v2-pro")
KNOWLEDGE_DIR = os.environ.get("HARVESTER_KNOWLEDGE_DIR", "knowledge")
PROMPT_PATH = os.environ.get("HARVESTER_PROMPT_PATH", str(SCRIPT_DIR.parent / "templates" / "harvest-prompt.md"))
# Where to look for API keys if not set via env
API_KEY_PATHS = [
os.path.expanduser("~/.config/nous/key"),
os.path.expanduser("~/.hermes/keymaxxing/active/minimax.key"),
os.path.expanduser("~/.config/openrouter/key"),
]
def find_api_key() -> str:
"""Find API key from common locations."""
for path in API_KEY_PATHS:
if os.path.exists(path):
with open(path) as f:
key = f.read().strip()
if key:
return key
return ""
def load_extraction_prompt() -> str:
"""Load the extraction prompt template."""
path = Path(PROMPT_PATH)
if not path.exists():
print(f"ERROR: Extraction prompt not found at {path}", file=sys.stderr)
print("Expected templates/harvest-prompt.md from issue #7", file=sys.stderr)
sys.exit(1)
return path.read_text(encoding='utf-8')
def call_llm(prompt: str, transcript: str, api_base: str, api_key: str, model: str) -> Optional[list[dict]]:
"""Call the LLM API to extract knowledge from a transcript."""
import urllib.request
messages = [
{"role": "system", "content": prompt},
{"role": "user", "content": f"Extract knowledge from this session transcript:\n\n{transcript}"}
]
payload = json.dumps({
"model": model,
"messages": messages,
"temperature": 0.1, # Low temp for consistent extraction
"max_tokens": 4096
}).encode('utf-8')
req = urllib.request.Request(
f"{api_base}/chat/completions",
data=payload,
headers={
"Authorization": f"Bearer {api_key}",
"Content-Type": "application/json"
},
method="POST"
)
try:
with urllib.request.urlopen(req, timeout=60) as resp:
result = json.loads(resp.read().decode('utf-8'))
content = result["choices"][0]["message"]["content"]
return parse_extraction_response(content)
except Exception as e:
print(f"ERROR: LLM API call failed: {e}", file=sys.stderr)
return None
def parse_extraction_response(content: str) -> Optional[list[dict]]:
"""Parse the LLM response to extract knowledge items.
Handles various response formats: raw JSON, markdown-wrapped JSON, etc.
"""
# Try direct JSON parse first
try:
data = json.loads(content)
if isinstance(data, dict) and 'knowledge' in data:
return data['knowledge']
if isinstance(data, list):
return data
except json.JSONDecodeError:
pass
# Try extracting JSON from markdown code blocks
import re
json_match = re.search(r'```(?:json)?\s*({.*?})\s*```', content, re.DOTALL)
if json_match:
try:
data = json.loads(json_match.group(1))
if isinstance(data, dict) and 'knowledge' in data:
return data['knowledge']
if isinstance(data, list):
return data
except json.JSONDecodeError:
pass
# Try finding any JSON object with knowledge array
json_match = re.search(r'({[^{}]*"knowledge"[^{}]*[[sS]*?][^{}]*})', content)
if json_match:
try:
data = json.loads(json_match.group(1))
return data.get('knowledge', [])
except json.JSONDecodeError:
pass
print(f"WARNING: Could not parse LLM response as JSON", file=sys.stderr)
print(f"Response preview: {content[:500]}", file=sys.stderr)
return None
def load_existing_knowledge(knowledge_dir: str) -> dict:
"""Load the existing knowledge index."""
index_path = Path(knowledge_dir) / "index.json"
if not index_path.exists():
return {"version": 1, "last_updated": "", "total_facts": 0, "facts": []}
try:
with open(index_path, 'r', encoding='utf-8') as f:
return json.load(f)
except (json.JSONDecodeError, IOError) as e:
print(f"WARNING: Could not load knowledge index: {e}", file=sys.stderr)
return {"version": 1, "last_updated": "", "total_facts": 0, "facts": []}
def fact_fingerprint(fact: dict) -> str:
"""Generate a deduplication fingerprint for a fact.
Uses the fact text normalized (lowercase, stripped) as the key.
Similar facts will have similar fingerprints.
"""
text = fact.get('fact', '').lower().strip()
# Normalize whitespace
text = ' '.join(text.split())
return hashlib.md5(text.encode('utf-8')).hexdigest()
def deduplicate(new_facts: list[dict], existing: list[dict], similarity_threshold: float = 0.8) -> list[dict]:
"""Remove duplicate facts from new_facts that already exist in the knowledge store.
Uses fingerprint matching for exact dedup and simple overlap check for near-dupes.
"""
existing_fingerprints = set()
existing_texts = []
for f in existing:
fp = fact_fingerprint(f)
existing_fingerprints.add(fp)
existing_texts.append(f.get('fact', '').lower().strip())
unique = []
for fact in new_facts:
fp = fact_fingerprint(fact)
if fp in existing_fingerprints:
continue
# Check for near-duplicates using simple word overlap
fact_words = set(fact.get('fact', '').lower().split())
is_dup = False
for existing_text in existing_texts:
existing_words = set(existing_text.split())
if not fact_words or not existing_words:
continue
overlap = len(fact_words & existing_words) / max(len(fact_words | existing_words), 1)
if overlap >= similarity_threshold:
is_dup = True
break
if not is_dup:
unique.append(fact)
existing_fingerprints.add(fp)
existing_texts.append(fact.get('fact', '').lower().strip())
return unique
def validate_fact(fact: dict) -> bool:
"""Validate a single knowledge item has required fields."""
required = ['fact', 'category', 'repo', 'confidence']
for field in required:
if field not in fact:
return False
if not isinstance(fact['fact'], str) or not fact['fact'].strip():
return False
valid_categories = ['fact', 'pitfall', 'pattern', 'tool-quirk', 'question']
if fact['category'] not in valid_categories:
return False
if not isinstance(fact.get('confidence', 0), (int, float)):
return False
if not (0.0 <= fact['confidence'] <= 1.0):
return False
return True
def write_knowledge(index: dict, new_facts: list[dict], knowledge_dir: str, source_session: str = ""):
"""Write new facts to the knowledge store."""
kdir = Path(knowledge_dir)
kdir.mkdir(parents=True, exist_ok=True)
# Add source tracking to each fact
for fact in new_facts:
fact['source_session'] = source_session
fact['harvested_at'] = datetime.now(timezone.utc).isoformat()
# Update index
index['facts'].extend(new_facts)
index['total_facts'] = len(index['facts'])
index['last_updated'] = datetime.now(timezone.utc).isoformat()
# Write index
index_path = kdir / "index.json"
with open(index_path, 'w', encoding='utf-8') as f:
json.dump(index, f, indent=2, ensure_ascii=False)
# Also write per-repo markdown files for human reading
repos = {}
for fact in new_facts:
repo = fact.get('repo', 'global')
repos.setdefault(repo, []).append(fact)
for repo, facts in repos.items():
if repo == 'global':
md_path = kdir / "global" / "harvested.md"
else:
md_path = kdir / "repos" / f"{repo}.md"
md_path.parent.mkdir(parents=True, exist_ok=True)
# Append to existing or create new
mode = 'a' if md_path.exists() else 'w'
with open(md_path, mode, encoding='utf-8') as f:
if mode == 'w':
f.write(f"# Knowledge: {repo}\n\n")
f.write(f"## Harvested {datetime.now(timezone.utc).strftime('%Y-%m-%d %H:%M')}\n\n")
for fact in facts:
icon = {'fact': '📋', 'pitfall': '⚠️', 'pattern': '🔄', 'tool-quirk': '🔧', 'question': ''}.get(fact['category'], '')
f.write(f"- {icon} **{fact['category']}** (conf: {fact['confidence']:.1f}): {fact['fact']}\n")
f.write("\n")
def harvest_session(session_path: str, knowledge_dir: str, api_base: str, api_key: str,
model: str, dry_run: bool = False, min_confidence: float = 0.3) -> dict:
"""Harvest knowledge from a single session.
Returns: dict with stats (facts_found, facts_new, facts_dup, elapsed_seconds, error)
"""
start_time = time.time()
stats = {
'session': session_path,
'facts_found': 0,
'facts_new': 0,
'facts_dup': 0,
'elapsed_seconds': 0,
'error': None
}
try:
# 1. Read session
messages = read_session(session_path)
if not messages:
stats['error'] = "Empty session file"
return stats
# 2. Extract conversation
conv = extract_conversation(messages)
if not conv:
stats['error'] = "No conversation turns found"
return stats
# 3. Truncate for context window
truncated = truncate_for_context(conv, head=50, tail=50)
transcript = messages_to_text(truncated)
# 4. Load extraction prompt
prompt = load_extraction_prompt()
# 5. Call LLM
raw_facts = call_llm(prompt, transcript, api_base, api_key, model)
if raw_facts is None:
stats['error'] = "LLM extraction failed"
return stats
# 6. Validate
valid_facts = [f for f in raw_facts if validate_fact(f) and f.get('confidence', 0) >= min_confidence]
stats['facts_found'] = len(valid_facts)
# 7. Deduplicate
existing_index = load_existing_knowledge(knowledge_dir)
existing_facts = existing_index.get('facts', [])
new_facts = deduplicate(valid_facts, existing_facts)
stats['facts_new'] = len(new_facts)
stats['facts_dup'] = len(valid_facts) - len(new_facts)
# 8. Write (unless dry run)
if new_facts and not dry_run:
write_knowledge(existing_index, new_facts, knowledge_dir, source_session=session_path)
stats['elapsed_seconds'] = round(time.time() - start_time, 2)
return stats
except Exception as e:
stats['error'] = str(e)
stats['elapsed_seconds'] = round(time.time() - start_time, 2)
return stats
def batch_harvest(sessions_dir: str, knowledge_dir: str, api_base: str, api_key: str,
model: str, since: str = "", limit: int = 0, dry_run: bool = False) -> list[dict]:
"""Harvest knowledge from multiple sessions in batch."""
sessions_path = Path(sessions_dir)
if not sessions_path.is_dir():
print(f"ERROR: Sessions directory not found: {sessions_dir}", file=sys.stderr)
return []
# Find session files
session_files = sorted(sessions_path.glob("*.jsonl"), reverse=True) # Newest first
# Filter by date if --since provided
if since:
since_dt = datetime.fromisoformat(since.replace('Z', '+00:00'))
filtered = []
for sf in session_files:
# Try to parse timestamp from filename (common format: session_YYYYMMDD_HHMMSS_hash.jsonl)
try:
parts = sf.stem.split('_')
if len(parts) >= 3:
date_str = parts[1]
file_dt = datetime.strptime(date_str, '%Y%m%d').replace(tzinfo=timezone.utc)
if file_dt >= since_dt:
filtered.append(sf)
except (ValueError, IndexError):
# If we can't parse the date, include the file (be permissive)
filtered.append(sf)
session_files = filtered
# Apply limit
if limit > 0:
session_files = session_files[:limit]
print(f"Harvesting {len(session_files)} sessions...")
results = []
for i, sf in enumerate(session_files, 1):
print(f"[{i}/{len(session_files)}] {sf.name}...", end=" ", flush=True)
stats = harvest_session(str(sf), knowledge_dir, api_base, api_key, model, dry_run)
if stats['error']:
print(f"ERROR: {stats['error']}")
else:
print(f"{stats['facts_new']} new, {stats['facts_dup']} dup ({stats['elapsed_seconds']}s)")
results.append(stats)
return results
def main():
parser = argparse.ArgumentParser(description="Harvest knowledge from session transcripts")
parser.add_argument('--session', help='Path to a single session JSONL file')
parser.add_argument('--batch', action='store_true', help='Batch mode: process multiple sessions')
parser.add_argument('--sessions-dir', default=os.path.expanduser('~/.hermes/sessions'),
help='Directory containing session files (default: ~/.hermes/sessions)')
parser.add_argument('--output', default='knowledge', help='Output directory for knowledge store')
parser.add_argument('--since', default='', help='Only process sessions after this date (YYYY-MM-DD)')
parser.add_argument('--limit', type=int, default=0, help='Max sessions to process (0=unlimited)')
parser.add_argument('--api-base', default=DEFAULT_API_BASE, help='LLM API base URL')
parser.add_argument('--api-key', default='', help='LLM API key (or set HARVESTER_API_KEY)')
parser.add_argument('--model', default=DEFAULT_MODEL, help='Model to use for extraction')
parser.add_argument('--dry-run', action='store_true', help='Preview without writing to knowledge store')
parser.add_argument('--min-confidence', type=float, default=0.3, help='Minimum confidence threshold')
args = parser.parse_args()
# Resolve API key
api_key = args.api_key or DEFAULT_API_KEY or find_api_key()
if not api_key:
print("ERROR: No API key found. Set HARVESTER_API_KEY or store in one of:", file=sys.stderr)
for p in API_KEY_PATHS:
print(f" {p}", file=sys.stderr)
sys.exit(1)
# Resolve knowledge directory
knowledge_dir = args.output
if not os.path.isabs(knowledge_dir):
knowledge_dir = os.path.join(SCRIPT_DIR.parent, knowledge_dir)
if args.session:
# Single session mode
stats = harvest_session(
args.session, knowledge_dir, args.api_base, api_key, args.model,
dry_run=args.dry_run, min_confidence=args.min_confidence
)
print(json.dumps(stats, indent=2))
if stats['error']:
sys.exit(1)
elif args.batch:
# Batch mode
results = batch_harvest(
args.sessions_dir, knowledge_dir, args.api_base, api_key, args.model,
since=args.since, limit=args.limit, dry_run=args.dry_run
)
total_new = sum(r['facts_new'] for r in results)
total_dup = sum(r['facts_dup'] for r in results)
errors = sum(1 for r in results if r['error'])
print(f"\nDone: {total_new} new facts, {total_dup} duplicates, {errors} errors")
else:
parser.print_help()
sys.exit(1)
if __name__ == '__main__':
main()

142
scripts/session_reader.py Normal file
View File

@@ -0,0 +1,142 @@
#!/usr/bin/env python3
"""
session_reader.py — Parse Hermes session JSONL transcripts.
Each line in a session file is a JSON object representing a message.
Standard fields: role (user|assistant|system), content (str), timestamp (str).
Tool calls and tool results are also captured.
"""
import json
import sys
from pathlib import Path
from typing import Iterator, Optional
def read_session(path: str) -> list[dict]:
"""Read a session JSONL file and return all messages as a list."""
messages = []
with open(path, 'r', encoding='utf-8') as f:
for line_num, line in enumerate(f, 1):
line = line.strip()
if not line:
continue
try:
msg = json.loads(line)
messages.append(msg)
except json.JSONDecodeError as e:
print(f"WARNING: Skipping malformed JSON at line {line_num}: {e}", file=sys.stderr)
return messages
def read_session_iter(path: str) -> Iterator[dict]:
"""Iterate over session messages without loading all into memory."""
with open(path, 'r', encoding='utf-8') as f:
for line_num, line in enumerate(f, 1):
line = line.strip()
if not line:
continue
try:
yield json.loads(line)
except json.JSONDecodeError as e:
print(f"WARNING: Skipping malformed JSON at line {line_num}: {e}", file=sys.stderr)
def extract_conversation(messages: list[dict]) -> list[dict]:
"""Extract user/assistant conversation turns, skipping tool-only messages."""
conversation = []
for msg in messages:
role = msg.get('role', '')
content = msg.get('content', '')
# Skip empty messages and pure tool calls
if role in ('user', 'assistant', 'system'):
if isinstance(content, str) and content.strip():
conversation.append({
'role': role,
'content': content.strip(),
'timestamp': msg.get('timestamp', '')
})
elif isinstance(content, list):
# Multimodal content — extract text parts
text_parts = []
for part in content:
if isinstance(part, dict) and part.get('type') == 'text':
text_parts.append(part.get('text', ''))
if text_parts:
conversation.append({
'role': role,
'content': '\n'.join(text_parts),
'timestamp': msg.get('timestamp', '')
})
return conversation
def truncate_for_context(messages: list[dict], head: int = 50, tail: int = 50) -> list[dict]:
"""Truncate long sessions: keep first N + last N messages.
This preserves session start (initial context) and end (final results),
skipping the messy middle of long debugging sessions.
"""
if len(messages) <= head + tail:
return messages
truncated = messages[:head]
truncated.append({
'role': 'system',
'content': f'[{len(messages) - head - tail} messages truncated]',
'timestamp': ''
})
truncated.extend(messages[-tail:])
return truncated
def messages_to_text(messages: list[dict]) -> str:
"""Convert message list to plain text for LLM consumption."""
lines = []
for msg in messages:
role = msg.get('role', 'unknown').upper()
content = msg.get('content', '')
if msg.get('role') == 'system' and 'truncated' in content:
lines.append(f'--- {content} ---')
else:
lines.append(f'{role}: {content}')
return '\n\n'.join(lines)
def get_session_metadata(path: str) -> dict:
"""Extract metadata from a session file (first message often has config info)."""
messages = read_session(path)
if not messages:
return {'path': path, 'message_count': 0}
first = messages[0]
last = messages[-1]
return {
'path': path,
'message_count': len(messages),
'first_timestamp': first.get('timestamp', ''),
'last_timestamp': last.get('timestamp', ''),
'first_role': first.get('role', ''),
'has_tool_calls': any(m.get('tool_calls') for m in messages),
}
if __name__ == '__main__':
if len(sys.argv) < 2:
print(f"Usage: {sys.argv[0]} <session.jsonl>")
sys.exit(1)
path = sys.argv[1]
meta = get_session_metadata(path)
print(json.dumps(meta, indent=2))
messages = read_session(path)
conv = extract_conversation(messages)
print(f"\nConversation: {len(conv)} turns")
truncated = truncate_for_context(conv)
print(f"After truncation: {len(truncated)} turns")
print(f"\nPreview (first 500 chars):")
print(messages_to_text(truncated[:5])[:500])

View File

@@ -1,38 +0,0 @@
#!/usr/bin/env python3
"""Validate knowledge files and index.json against the schema."""
import json, sys
from pathlib import Path
VALID_CATEGORIES = {"fact", "pitfall", "pattern", "tool-quirk", "question"}
REQUIRED = {"id", "fact", "category", "domain", "confidence"}
def validate_fact(fact, src=""):
errs = []
for f in REQUIRED:
if f not in fact: errs.append(f"{src}: missing '{f}'")
if "category" in fact and fact["category"] not in VALID_CATEGORIES:
errs.append(f"{src}: invalid category '{fact['category']}'")
if "confidence" in fact:
if not isinstance(fact["confidence"], (int, float)) or not (0 <= fact["confidence"] <= 1):
errs.append(f"{src}: confidence must be 0.0-1.0")
if "id" in fact:
parts = fact["id"].split(":")
if len(parts) != 3: errs.append(f"{src}: id must be domain:category:sequence")
return errs
def main():
idx = Path(__file__).parent.parent / "knowledge" / "index.json"
if not idx.exists(): print(f"FAILED: {idx} not found"); sys.exit(1)
data = json.load(open(idx))
errs = []
seen = set()
for i, f in enumerate(data.get("facts", [])):
errs.extend(validate_fact(f, f"[{i}]"))
if "id" in f:
if f["id"] in seen: errs.append(f"duplicate id '{f['id']}'")
seen.add(f["id"])
if errs:
print(f"FAILED - {len(errs)} errors:"); [print(f" x {e}") for e in errs]; sys.exit(1)
print(f"PASSED - {len(data.get('facts', []))} facts")
if __name__ == "__main__": main()