GitHub repository access in containerized environments using REST API and credential detection. Use when git clone fails, or when accessing private repos/writing files via API.
日本語の概要は準備中です。原文の説明を表示しています。
Puts oaustegard/muninn-utilities on the container Python path — runs scripts/boot.sh, which shallow-clones the repo (codeload tarball as fallback) and writes the .pth. The boot payload itself comes from the Muninn MCP connector: call its `boot` tool after this script. Trigger on the first message of any new Muninn conversation, or when the user says "boot up" / "boot" — unless the message contains "skip boot". Idempotent: a warm container exits in ~0s via a sentinel.
インストールする前に、エージェントに与えられる指示の中身を確認できます。
Two steps, in this order:
bash "$(dirname "$(find / -name boot.sh -path '*muninn-boot*' 2>/dev/null | head -1)")/boot.sh"
then call the Muninn MCP boot tool. The script fetches code; the tool
delivers identity, ops, recent memories and reminders. Memory reads and writes
go through the connector's recall / remember / forget / muninn_config
tools, so the container holds no Turso credential and this skill reads no
Turso.env. Direction in memory ce3b8b75, the plan in 69f3301c.
Prior to 2.0.0 this skill sideloaded claude-skills and ran the Python boot()
itself. Marketplace sync now places claude-skills in every session
(/root/.claude/skills/synced/<id>/ in Cowork, /mnt/skills/user on
Claude.ai), and the worker runs boot. "skinny boot" (BOOT_MODE=perch) has no
effect until the worker's boot takes a mode.
The boot tool's payload is larger than the harness output cap, so it comes
in parts. Call boot with no arguments; its footer says whether the payload is
complete or names the next part (boot({part: 2}), then 3, and so on). Call
every part the footer names before replying. The harness may persist a part to
a file instead of showing it ("Output too large ... saved to ..."); read that
file in full. A part that was persisted has already been delivered, so never
call boot again with the same part to recover it.
Check metadata.version at the top of this file right before running. Below
2.0.0 the on-disk script is the old one that sideloads claude-skills and runs
the Python boot() itself; marketplace sync can replace the copy mid-session,
so a session may run 1.x once and 2.x next.
/home/claude/.muninn-booted present and
/home/claude/muninn-utilities/muninn_utils exists → exit 0.$MUNINN_PROJECT_DIR/GitHub.env (default /mnt/project) if present.
Transitional — it goes once GitHub write tools exist on the worker. The
container presets GH_TOKEN=proxy-injected, a truthy placeholder; without
the file, GitHub API calls from the container 401 and the script says so.
Memory tools are unaffected.git clone --depth 1 oaustegard/muninn-utilities at MUNINN_UTILS_REF
(default main; branch, tag or full SHA) → /home/claude/muninn-utilities.
Measured 1.3s cold in Cowork. Fallback where the git proxy refuses
github.com: the codeload tarball, which works on Claude.ai and is 403 in
Cowork (there git works). If both fail the script exits 1.muninn-remembering.pth at a site-packages directory resolved from
sys.path at runtime (/root/.local/lib/python3.11/site-packages in
Cowork, measured 2026-09-19) listing the repo, remembering/, $HOME, and
every skill scripts/ directory found under either skills root.boot tool.Only GitHub.env matters now, and only for GitHub API work from the container.
project_read it and write it to /mnt/project/GitHub.env with the Write tool
(a bash heredoc carrying a secret trips the permission classifier), or skip it
when the task touches no GitHub API.
MUNINN_UTILS_REF defaults to main; the point of cloning is that changes land
on the next cold boot with no version bump. Pass a SHA to test an unmerged
state. rm /home/claude/.muninn-booted forces a re-pull in a warm container.
This file and boot.sh are the one thing that cannot self-update: the running
copy is whatever the session mounted (project upload or marketplace sync), so
a change here needs a push to both muninn-utilities/muninn-boot and the
claude-skills/muninn-boot mirror, with metadata.version bumped.
まだレビューはありません。使ってみた感想をお寄せください。
概要と使いどころ
GitHub repository access in containerized environments using REST API and credential detection. Use when git clone fails, or when accessing private repos/writing files via API.
日本語の概要は準備中です。原文の説明を表示しています。
Decide which model, effort level, and cascade shape each subagent gets, and how to keep improvement loops safe (evaluator-as-selector, stop on regression). Routes on measured cost-per-completed-task rather than per-token price, because a tier's token count varies more by task shape than price varies across tiers. Covers per-model effort semantics, the concision lever, cascade preconditions, context handoff, and watching a subagent fan-out live. Use when choosing a subagent's model or effort, when choosing how to escalate a failed attempt, or when asked which model or effort a task should get. Grounded in a 2026-07 calibration, a 2026-08 coding-cost study, a 2026-09 seeded-bug repair battery, a 2026-10 Haiku 5.5 ladder priced per spawn, and a 2026-10 retry-versus-escalation ladder on 295 SWE-bench Verified tasks; Managed Agents API specifics are operational, not calibrated.
日本語の概要は準備中です。原文の説明を表示しています。
Securely manages API credentials for multiple providers (Anthropic Claude, Google Gemini, GitHub). Use when skills need to access stored API keys for external service invocations.
日本語の概要は準備中です。原文の説明を表示しています。
Guidance for asking clarifying questions when user requests are ambiguous, have multiple valid approaches, or require critical decisions. Use when implementation choices exist that could significantly affect outcomes.
日本語の概要は準備中です。原文の説明を表示しています。
Pre-change blast-radius report for a symbol or file. Walks tree-sitting references, augments with a plain-text scan over non-parsed files (configs, plain docs), and clusters affected sites by feature (`_FEATURES.md`) or top-level package. Use when about to refactor, rename, or delete something in a repo you don't own — "what breaks if I change `validateUser`", "who calls this", "is this safe to remove", "where is this used", "blast radius", "impact analysis". This is the CONVERGENT pre-change risk skill — for "what is this repo?" use exploring-codebases; for "where is X?" use searching-codebases.
日本語の概要は準備中です。原文の説明を表示しています。
Read Bluesky/ATProto without depending on Bluesky's AppView — interactions on a user's posts, thread replies, any repo's records, network-wide backlinks across every lexicon, an account's handle and PDS history, which collections are active network-wide, and layer-by-layer outage diagnosis. Use when bsky.app or the AppView is down or slow, when a Bluesky read returns timeouts or 5xx, when asked who liked/replied/quoted/reposted something, when pulling live Bluesky context cheaply, when reading records straight from a PDS, when asked who references a record or account anywhere on the network, whether an account has changed handle or migrated servers, or which lexicons and non-Bluesky apps are active on atproto. Complements browsing-bluesky, which routes everything through the AppView and fails when it does.
日本語の概要は準備中です。原文の説明を表示しています。