CorvidLabs Let Discovery
Let is a host-neutral locator and workbed for agent assets. Discovery commands
(doctor, where, find, show, open, context, history, skill routing) are
read-oriented. Workbed commands (init, worktree add|remove, memory set|delete,
super init-example) can write under project or user .let paths—do not use them unless
the task authorizes local writes.
Let does not send prompts to an agent or expose live model reasoning.
Install
Let is a Fledge plugin. Prefer the Fledge entrypoint (the shell builtin let is unrelated):
fledge plugins install CorvidLabs/let
fledge let --help
Two modes
Public-only (default when publishing or working from public evidence)
Do not invoke Let. Discovery can surface local host metadata outside the public
repository (doctor lists host homes; project-scope finds still pull user catalogs for
some kinds by default). Never rely on filtering or redaction after a broad query.
Stay inside the confirmed repository:
fledge work status
fledge run --list
git status --short --branch
git ls-files
Report the Let step as blocked by the current isolation boundary and continue with repository-local Git and Fledge facts.
Authorized local metadata
Use Let only when the task explicitly permits discovery of local agent metadata. Inspect help, then choose the narrowest query. Prefer exact identifiers over a full context pack.
fledge let doctor --json
fledge let where .
fledge let find worktrees --scope project --json
fledge let find sessions --scope project --json
fledge let find skills --query <text> --json
fledge let find agents --json
fledge let skill route "<request>" --json
fledge let show skill <id-or-name> --json
fledge let open <path> --json
fledge let context --pack brief --json
fledge let history --scope project --json
Kinds include: instructions, skills, agents, commands, worktrees, sessions,
tasks, memory, mcp, plugins, workflows, superskills.
Scopes: project | user | all. Start with project when authorized; widen only with
cause. history defaults to user if --scope is omitted—always pass
--scope project for repo-bounded history.
Project-scope privacy boundary
Check the installed tool's scope semantics and configuration before discovery.
--scope project alone is not proof of isolation; federated discovery can behave as follows:
| Safer under project scope | Still can pull user-global / sibling context |
|---|---|
sessions, memory, tasks (repo-bound by design) |
skills, agents, commands, workflows (user catalogs included by default via find.include_user_skills) |
instructions may include user-level CLAUDE.md / AGENTS.md |
|
worktrees / where can list sibling worktrees for the same repo |
Treat every returned path and session identifier as private unless public provenance is
independently established. context does not include sessions—call find sessions
explicitly when needed. Session and memory cards are path-oriented, not full transcripts.
Optional local dashboard (still local metadata; not for public-only work):
fledge let web
Freshness and limits
- Session records are activity hints, not proof of completed work.
- Compare session timestamps with the worktree, commits, pull request, CI, and sandbox.
- A missing session does not prove no work exists; discovery only indexes configured hosts.
- Let cannot bypass a provider connection or message an agent.
When intervention is needed, pass the confirmed session and worktree to the rune skill.
For the full coordination sequence, use agent-coordination.