AgentSkillsCN

jarvis-settings

查看并更新Jarvis配置。当用户说“/jarvis-settings”、“Jarvis设置”、“更新配置”、“更改存储库路径”或“配置自动提取”时使用。

SKILL.md
--- frontmatter
name: jarvis-settings
description: View and update Jarvis configuration. Use when user says "/jarvis-settings", "Jarvis settings", "update config", "change vault path", or "configure auto-extract".
user_invocable: true

/jarvis-settings - Jarvis Configuration

Re-runnable configuration manager for Jarvis. Jump to any setting, change what you need, leave the rest.

First-time setup? If ~/.jarvis/config.json doesn't exist, guide the user through initial config (vault path + basics), then offer to explore advanced settings.

Execution Flow

1. Load and display current config

Read ~/.jarvis/config.json and show a summary:

code
Jarvis Configuration
--------------------
Vault:        ~/.jarvis/vault/
File Format:  Markdown (.md)
Auto-Extract: background (min text: 200 chars)
Memory DB:    ~/.jarvis/db/ (configurable)
Shell:        jarvis command in PATH
Version:      1.19.0

If config doesn't exist, say: "No config found. Let's set up the basics." and go to Step 2a (first-time flow).

2. Ask what to configure

Use AskUserQuestion:

code
AskUserQuestion:
  questions:
    - question: "What would you like to configure?"
      header: "Settings"
      options:
        - label: "Vault path"
          description: "Change where Jarvis stores your knowledge"
        - label: "File format"
          description: "Choose Markdown (.md) or Org-mode (.org) for new files"
        - label: "Auto-Extract"
          description: "Configure observation capture from conversations"
        - label: "Advanced settings"
          description: "Promotion thresholds, vault paths, memory tuning"
      multiSelect: false

2a. First-time flow (no config exists)

If no config exists, skip the menu and walk through essentials:

  1. Vault path (Step 3a)
  2. File format (Step 3b) - Markdown or Org-mode
  3. Auto-Extract mode (Step 3c) - quick preset selection only
  4. Shell integration (Step 3f below)
  5. Write full config with ALL defaults visible
  6. Offer: "Want to explore advanced settings? Or you're good to go."

3a. Vault path

Scan for common vault locations, then ask:

code
AskUserQuestion:
  questions:
    - question: "Where should Jarvis store your knowledge vault?"
      header: "Vault"
      options:
        - label: "~/.jarvis/vault/ (Recommended)"
          description: "Starter vault - good for trying Jarvis or if you don't have a PKM"
        - label: "[detected Obsidian vault]"
          description: "Use your existing Obsidian vault at [path]"
        - label: "Enter custom path"
          description: "Specify your own directory"
      multiSelect: false

Scan these locations for Obsidian vaults (contain .obsidian/):

  • ~/Documents/Obsidian*
  • ~/Obsidian*
  • ~/vaults/*
  • ~/Library/Mobile Documents/iCloud~md~obsidian/Documents/*

What is a vault? If the user seems confused, explain briefly:

A vault is just a folder of markdown files. If you use Obsidian, it's your Obsidian vault. If not, Jarvis creates a simple folder structure for journals, notes, and strategic context.

After changing vault path: ask if user wants to reindex for semantic search.

3b. File format

Show current format (file_format in config, default: md), then ask:

code
AskUserQuestion:
  questions:
    - question: "File format for new vault files? (current: [format])"
      header: "Format"
      options:
        - label: "Markdown (.md) (Recommended)"
          description: "Standard format, works with Obsidian and most tools"
        - label: "Org-mode (.org)"
          description: "For Emacs/Org-mode users. Existing .md files remain readable."
      multiSelect: false

Map to config values:

  • "Markdown" -> "md"
  • "Org-mode" -> "org"

Update file_format in config. Note: this only affects new files. Existing files in either format are always readable and searchable.

3c. Auto-Extract

Show current mode, then ask:

code
AskUserQuestion:
  questions:
    - question: "Auto-Extract mode? (current: [mode])"
      header: "Auto-Extract"
      options:
        - label: "Background (Recommended)"
          description: "Smart fallback: tries API first, then CLI. Near-zero cost (~$0.02/session)."
        - label: "Background API only"
          description: "Requires ANTHROPIC_API_KEY. Fastest extraction (~200ms)."
        - label: "Background CLI only"
          description: "Uses Claude CLI via OAuth. No API key needed (~2-5s)."
        - label: "Disabled"
          description: "No automatic observation capture."
      multiSelect: false

Map to mode values:

  • "Background" -> "background"
  • "Background API only" -> "background-api"
  • "Background CLI only" -> "background-cli"
  • "Disabled" -> "disabled"

If NOT disabled, ask about thresholds:

code
AskUserQuestion:
  questions:
    - question: "Extraction sensitivity?"
      header: "Tuning"
      options:
        - label: "Defaults (Recommended)"
          description: "Min text: 200 chars per turn"
        - label: "Frequent capture"
          description: "Min text: 100 chars (more observations, slightly noisier)"
        - label: "Conservative"
          description: "Min text: 500 chars (fewer, higher quality)"
        - label: "Custom values"
          description: "Enter your own min_turn_chars threshold"
      multiSelect: false

Preset mapping:

  • "Defaults" -> min_turn_chars: 200
  • "Frequent capture" -> min_turn_chars: 100
  • "Conservative" -> min_turn_chars: 500
  • "Custom" -> Ask for min_turn_chars (int)

3d. Advanced settings

code
AskUserQuestion:
  questions:
    - question: "Which advanced area?"
      header: "Advanced"
      options:
        - label: "Promotion thresholds"
          description: "When tier 2 content gets promoted to vault files"
        - label: "Vault directory paths"
          description: "Where journals, notes, inbox, etc. are stored"
        - label: "Memory system"
          description: "Secret detection, importance scoring, DB location"
        - label: "Per-prompt search"
          description: "Automatic vault memory injection on every message"
        - label: "Back to main menu"
      multiSelect: false

Promotion thresholds

Show current values and let user adjust:

SettingDefaultDescription
importance_threshold0.85Auto-promote score threshold
retrieval_count_threshold3Promote after N retrievals
age_importance_days30Age-based promotion trigger
age_importance_score0.7Importance threshold for aged content
on_promoted_file_deleted"remove"What happens when promoted file is deleted

Always offer "Reset to defaults" as an option.

Vault directory paths

Show all vault-relative paths with current vs default:

Path NameDefaultCurrent
journal_jarvisjournal/jarvis[current or default]
journal_dailyjournal/daily...
notesnotes...
workwork...
inboxinbox...
inbox_todoistinbox/todoist...
templatestemplates...
strategic.jarvis/strategic...
observations_promoted.jarvis/memories/observations...
patterns_promoted.jarvis/memories/patterns...
learnings_promoted.jarvis/memories/learnings...
decisions_promoted.jarvis/memories/decisions...

Let user change specific paths. All paths are relative to vault root.

Memory system

SettingDefaultDescription
secret_detectiontrueScan content for secrets before storing
importance_scoringtrueScore content importance on write
recency_boost_days7Boost recent content in search results
default_importance0.5Default importance score (0.0–1.0) for new content
chroma_hostlocalhostChromaDB server hostname
chroma_port8743ChromaDB server port
chroma_sslfalseUse HTTPS for ChromaDB connection
chroma_data_path~/.jarvis/dbChromaDB server data directory

Cross-encoder reranking

SettingDefaultDescription
enabledtrueEnable cross-encoder reranking for /jarvis-recall results
candidate_count100How many results to over-fetch for reranking
top_k10Final result count after reranking
alpha0.7Blend weight (0.0=vector only, 1.0=reranker only)
max_latency_ms1000Latency budget before fallback to vector scores
batch_size32Tokenization batch size for ONNX inference

Config key: memory.reranking

Note: The ONNX model (~23MB) is downloaded automatically on first use to ~/.jarvis/models/cross-encoder/. Not applied to per-prompt search.

Per-prompt search

SettingDefaultDescription
enabledtrueMaster switch for automatic memory recall
threshold0.5Minimum relevance score (0.4=aggressive, 0.5=balanced, 0.6=conservative)
budget8000Total character budget for injection (split 50/50 between tier2 and vault)

Offer presets:

  • "Balanced" (default) -> threshold: 0.5, budget: 8000
  • "Aggressive recall" -> threshold: 0.4, budget: 12000
  • "Conservative" -> threshold: 0.6, budget: 4000
  • "Disabled" -> enabled: false
  • "Custom" -> Ask for each setting individually

Config key: memory.per_prompt_search

3e. View full config

Pretty-print ~/.jarvis/config.json grouped by section:

code
=== Core ===
vault_path:      ~/.jarvis/vault/
file_format:     md
vault_confirmed: true
configured_at:   2026-02-08T10:30:00Z

=== Memory ===
chroma_host:          localhost
chroma_port:          8743
chroma_data_path:     ~/.jarvis/db
secret_detection:     true
importance_scoring:   true
recency_boost_days:   7
default_importance:   0.5

=== Per-Prompt Search ===
enabled:              true
threshold:            0.5
budget:               8000

=== Auto-Extract ===
mode:                 background
min_turn_chars:       200
max_transcript_lines: 500
debug:                false

=== Promotion ===
importance_threshold:        0.85
retrieval_count_threshold:   3
age_importance_days:         30
age_importance_score:        0.7
on_promoted_file_deleted:    remove

=== Vault Paths (relative to vault root) ===
journal_jarvis:         journal/jarvis
journal_daily:          journal/daily
notes:                  notes
...

Highlight any non-default values with a marker. After viewing, return to main menu.

3f. Install jarvis command

Check if the jarvis executable is already installed and in PATH (command -v jarvis).

If not installed or outdated, find the shell script in the plugin distribution and install it:

  1. Locate jarvis.sh in the plugin's shell/ directory (use the skill's base directory, two levels up from skills/<name>/)
  2. Determine install directory:
    • ~/.local/bin/ if it exists and is in PATH (preferred)
    • /usr/local/bin/ if writable
    • Fallback: ~/.local/bin/ (create it, warn about PATH)
  3. Copy to <install_dir>/jarvis and chmod +x
  4. If directory not in PATH, tell user to add it: export PATH="~/.local/bin:$PATH"

If old shell function markers exist in RC files (# Jarvis AI Assistant START in ~/.zshrc, ~/.bashrc, or ~/.bash_profile), offer to clean them up by removing the START/END block.

4. Write config

Merge changes into existing config. Never overwrite keys that weren't changed.

For first-time setup, read the default config template shipped with the plugin:

  1. Locate template: <plugin_root>/defaults/config.json
    • Use the skill's base directory (two levels up from skills/<name>/) to find defaults/
    • Or read it via: Read tool on the defaults/config.json file relative to the plugin root
  2. Read template, substitute user values:
    • vault_path -> user's chosen path
    • vault_confirmed -> true
    • configured_at -> current ISO 8601 timestamp
    • file_format -> user's chosen format ("md" or "org")
    • auto_extract.mode -> user's chosen mode
  3. Write to ~/.jarvis/config.json

For existing config updates, merge changes — don't overwrite keys that weren't changed.

Also ensure directories exist:

bash
mkdir -p [vault_path]
mkdir -p ~/.jarvis/db

5. Summary

Show what changed:

code
Settings updated!

Changed:
  - vault_path: ~/old/path -> ~/new/path
  - auto_extract.mode: disabled -> background

Quick Start:
  /jarvis-recall <query>  - Search vault semantically
  /jarvis-promote         - Review & promote observations
  /jarvis-settings    - Update configuration anytime
  $ jarvis            - Launch from terminal (if shell configured)

Key Rules

  • Menu-driven - User jumps to any section, not forced through a linear flow
  • Merge, don't overwrite - Preserve existing config keys when updating a section
  • Show defaults - First-time config includes ALL keys so users can discover options
  • No broken references - Don't mention skills or features that don't exist
  • Reindex offer - When vault path changes, offer to reindex for semantic search
  • Always offer "back" - User can return to main menu from any sub-section