mcpField
T
trylle-cli

trylle-cli

github.com/trylle-labsDevelopment0132 installsUpdated 1w ago
Repository

About trylle-cli

A skill guide for using `trylle-cli` to manage Trylle platform resources. It helps developers handle repositories, PRs, issues, CI, and cloud coding agents, with guidance on JSON output, safe operations, and credentials.

Key features

  1. 01Manage repositories, PRs, and issues
  2. 02Inspect CI status and logs
  3. 03Start and follow cloud coding agents
  4. 04Use JSON command output
  5. 05Follow operation and credential safeguards

Use cases

  1. 01Check repository context and status
  2. 02Create or review pull requests
  3. 03Investigate CI runs
  4. 04Manage cloud agent sessions

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/trylle-labs/skills --skill trylle-cli

After installing, type/trylle-cliin Claude Code to confirm it loaded.

Bundled files

References and scripts installed along with SKILL.md.

TERMS_OF_USE
agents/
openai.yaml
references/
command-catalog.md
inline-reviews.md
workflows.md

Details

From the project README

Trylle CLI, Skill Guide

Overview

Use try as the primary interface for Trylle platform operations. Use git for local version-control work, but do not substitute GitHub-specific tooling for Trylle resources.

By invoking any script in this skill, you agree to the Terms of Use in TERMS_OF_USE.

Workflow

1. Locate or install the binary

Use a local trylle-cli or try binary if possible. If it is not on PATH,

npm i -g @trylle/trylle-cli

One-time commands can be run via npx trylle-cli, example: npx trylle-cli version. The rest of documentation will use try as a reference.

Other ways to get the binary:

# Python Way
uv tool install trylle-cli

# Rust Way (slow and will build from source)
cargo install trylle-cli

2. Core usage

Important: CLI is tailored for humans to use, so you should add --json flag for agentic usage of commands.

  1. Check authentication with try auth status --json. Its kind is cli_key for a person's key or bot_token for a run-scoped token (automation run, CI job, Magic Task, or Magic Session), and botToken then shows the token's scopes, repositories, and expiry. Never run try auth token --show or print tokens unless the user explicitly requests the secret material and the environment can handle it safely.
  2. Resolve context with try repo context --json inside a checkout. Outside a checkout, pass -R OWNER/NAME to commands that support it or pass OWNER/NAME positionally to try repo view, clone, and similar repository commands.
  3. Before a write, run try <group> <command> --help, validate identifiers and the intended scope, then execute only the requested mutation.
  4. Re-read the affected resource. Return a web URL with try browse repo|pr|issue when it helps the user continue.

Route the task

  • Repositories and daily context: use try status for current user's summary and try repo.
  • Pull requests and reviews: use try pr.
  • Issue triage: use try issue.
  • Dependent branches and pull requests: use try stack.
  • CI status, logs, watching, and re-runs: use try ci.
  • Cloud coding agents: use try task to start or follow up agents and try session for their workspaces.
  • The Trylle app around the session you run in (local or cloud): use try app. try app open changes|files|terminal|pr shows a tab of the session's panel, try app open --pr <number|url|OWNER/NAME#N> shows a pull request there, and try app pr attach|detach|list tracks pull requests with the session. A PR you create with try pr create inside a session is attached already. Show the reader something when it helps them follow the work, not after every step.
  • Repository or organization configuration: use try actions or try org.
  • Accounts and operations: use try profile, try ssh-key, try automation, or try bot.
  • Resource links and uncovered public endpoints: use try browse or try api (try api --list [FILTER] --json lists every public endpoint with its parameters).
  • Local AI and review helpers: use try ai, try explain, try draft, or try diff.

Read references/command-catalog.md for the complete command map and exact argument patterns. Read references/workflows.md when creating or publishing a repository, opening or reviewing a pull request, triaging issues, handling stacks, diagnosing CI, or changing secrets and settings. Read references/inline-reviews.md before an automation submits or manages inline pull request review comments.

Safety and output rules

  • Treat repository deletion, PR merging or closing, issue closing, label replacement, key deletion, automation toggles, bot installations, repository renames or transfers, Magic Session deletion or force-pause, and secret changes as consequential operations. Show the exact target and inspect current state first.
  • Keep credentials out of command literals, logs, files, commit messages, and chat. Use pre-existing environment variables for secret values and unset temporary variables afterward.
  • Prefer --body-file to complex inline PR, review, issue, or comment bodies. Attach screenshots with --attach FILE[#ALT] (see workflows).
  • Use --json whenever available. Human output is intended for terminal reading and may change formatting.
  • Do not guess unsupported flags. The installed CLI's --help output is the source of truth.
  • Do not use try api when a first-class command exists. For raw calls, use only the public /v1 API path needed for the task.

Installing or update the skill

If some of the skill files are missing or CLI reference map is outdated, use commands to update the skill:

# add
npx skills add trylle-labs/skills

# update
npx skills update trylle-labs/skills --skill trylle-cli

Or even using uv:

uvx upd-skill -g trylle-labs/trylle-cli

Current CLI gotchas

  • Repository inference depends on a recognizable git remote. Pass -R OWNER/NAME explicitly when inference is ambiguous.
  • try pr current --json exits nonzero when the current branch has no open PR instead of returning JSON null. Use try repo context --json when absence is an expected state.
  • try pr review requires exactly one of --approve, --request-changes, or --comment.
  • try pr review also requires --commit-id when --comments or --comments-file submits inline comments.
  • try issue label --set replaces the complete label set; --add and --remove preserve unrelated labels.
  • try stack submit pushes the current branch to origin before creating the pull request.
  • try ci watch polls until the run finishes. CLI 1.1.6 and earlier returned a single log snapshot instead.
  • try repo delete and try repo transfer require --yes when not interactive, and try session delete does too.
  • Bot tokens cannot be minted or listed from the CLI: the platform injects a run-scoped TRYLLE_BOT_TOKEN. Bot management (try bot ...) requires a person's CLI key, and only a bot's creator can try bot edit it; --attach uploads also work with a bot token that has a comment or write scope.
  • try session logs reads session history, not a live shell; use try session ssh for a shell. Output is chronological and complete from the start of the session.
  • try app works only inside a Trylle session (it reads TRYLLE_SESSION_ID and TRYLLE_DESKTOP_ENDPOINT); elsewhere it says there is no session to act on. A panel request shows at once when the reader is on that session and otherwise waits until they open it. A cloud session tracks Trylle pull requests only.
  • List filters such as PR author and issue labels may be applied to the returned page. Increase the page size or paginate before concluding that no matching item exists.

Comments

Markdown supported