Skip to content

usage.md

Latest commit

 

History

History
240 lines (177 loc) · 7.06 KB

usage.md

File metadata and controls

240 lines (177 loc) · 7.06 KB
Raw

Usage

Install

fledge plugin (CorvidLabs projects)

fledge plugins install CorvidLabs/let
fledge let find worktrees --json
fledge let where .
fledge let skill route "find worktrees" --json

Uses the same fledge-v1 plugin protocol as other CorvidLabs plugins. Default --cwd is the fledge project root when omitted.

Standalone CLI

git clone https://github.com/CorvidLabs/let
cd let && bun install && bun link
let doctor --json
# or: ./bin/let doctor --json

The let standard

let is the host-neutral catalog for coding-agent assets. It does not move host files into .let/; it indexes them in place and surfaces one card shape for every kind.

Principle Meaning
Federation over relocation Host paths stay put
Closed kinds One enum for all hosts
Progressive disclosure find = cards; show = body or metadata
agent.3md first-class 3md agents + skill planes are native
Sessions/memory safe Path-only; never dump transcripts

Commands

All commands accept --json (default machine path for agents).

let web

Start the local, read-only Fleet dashboard.

let web
let web --port 8732
fledge let web

The dashboard binds only to 127.0.0.1 (default port 8731) and refreshes from Let's structured index cards. It puts agents first, with projects as a secondary view. Expand an agent to inspect its redacted task context, command, latest prompt or update, timestamps, and recent session output when locally available. It never provides controls, shell execution, or unredacted secrets.

let history (alias: usage)

Rank hosts and projects by session activity (path-only, no transcripts).

let history --json                    # default scope=user (this Mac)
let history --scope project --json    # current repo only

let mcp serve

Read-only MCP server on stdio (JSON-RPC 2.0). Tools: let_find, let_where, let_context, let_show, let_doctor, let_route, let_history.

let mcp serve

Workbed

let init --json
let worktree list --json
let worktree add my-feature --json
let memory set note '{"hello":true}' --json
let memory list --json
let super init-example --json
let super list --json
let config show --json

let doctor

Host roots, git/bun, config trust, agent.3md presence.

let doctor --json

let find <kind>

Federated index cards.

let find worktrees --json
let find skills --query worktree --json
let find skills --host agent3md --json
let find agents --json
let find commands --json
let find memory --json
let find sessions --json            # path-only cards, never bodies
let find plugins --json
let find mcp --json
let find tasks --json
let find workflows --json
let find superskills --json
let find instructions --json

Kinds: instructions, skills, agents, commands, worktrees, sessions, tasks, memory, mcp, plugins, workflows, superskills.

Hosts (--host): claude, grok, codex, cursor, openai, gemini, kimi, agent3md, git, project, let, …

Host × kind matrix

Host Worktrees Skills Agents Commands Sessions Memory Plugins MCP Tasks
claude yes yes yes yes path-only — yes settings/.mcp path-only
grok worktrees.db yes bundled — path-only memtrace — — —
codex yes optional yes — path-only sqlite/dir yes config path —
cursor weak yes yes yes chats path — — mcp.json plans
openai — yes — — — — — — —
gemini — — antigravity — history path brain/knowledge — mcp_config —
kimi — — config path — workspaces path user-history — — —
agent3md — skill planes agent.3md — — — — — —
let .let/worktrees .let/skills .let/agents — .let/sessions .let/memory — — superskills

Scopes: --scope project|user|all (default project). Project scope still includes user skill/agent catalogs when include_user_skills is true (default).

let where [path]

Classify cwd or path; sibling worktrees + instructions. related.sessions is always [] in v0.

let where .
let where /path/to/.claude/worktrees/wf_1 --json

let context

Brief or full pack. Never includes sessions.

let context --json
let context --pack full --cwd /path/to/repo --json

let show <kind> <id|name>

Progressive disclosure: load the body (or metadata) for one card.

let show skill find-worktrees --json
let show agent let --json
let show instructions AGENTS.md --json
let show worktree <id> --json
let show memory <id> --json          # metadata only
let show sessions <id> --json        # metadata only
  • skills / instructions / agents: full text body (agent.3md skills = plane body)
  • agent.3md agents: identity body + skill catalog payload (not every plane body)
  • worktrees: enrichment only (branch, head, exists)
  • sessions / memory / most tasks: metadata only
  • cursor plans (tasks): markdown body allowed
  • Ambiguous names return conflict with candidate ids

let open <path>

Classify a filesystem path + small preview (8 KiB). Refuses session jsonl bodies.

let open ./agent.3md --json
let open ./AGENTS.md --json

let skill route <text> (alias: let route)

Rank skills for a natural-language request. Uses agent.3md Agent.route when possible, then trigger phrases, then name/description fallback.

let skill route "find worktrees for this repo" --json
let route "where am i" --host agent3md --json
let route "list memory" --json

Each hit includes score, hits (matched phrases), skill card, tool template, and show (command to load the body).

agent.3md / 3md (let standard)

let depends on @corvidlabs/agent3md (canonical 3md parser + agent loader).

Discovered paths:

  • Project: agent.3md, root *.3md, agents/*.3md, .let/agents/*.3md
  • User: ~/.let/agent.3md, ~/.let/agents/*.3md, and *.3md under host agent dirs

Behavior:

  • find agents → agent documents (host=agent3md, meta.format=agent.3md)
  • find skills → skill planes as progressive cards (meta.z, triggers, tool)
  • show agent <name> → identity body + skill catalog payload
  • show skill <name> → single plane body
  • skill route → prefers agent.3md Agent.route, then triggers

This repo ships agent.3md describing let itself (dogfood).

Library

import { buildScanContext, findAssets, whereAmI, routeSkills } from "@corvidlabs/let";

const ctx = buildScanContext({ cwd: process.cwd() });
const trees = await findAssets("worktrees", ctx);
const memory = await findAssets("memory", ctx, { host: "grok" });
const here = whereAmI(ctx);
const ranked = await routeSkills("find skills about worktrees", ctx);

Quality gate

fledge lanes run verify