Agent Native

commit

The commit skill creates one local Git commit with a complete repository-wide diff, validating changes before staging and committing.

by NebiusOfficial publisherNot reviewed

Identity proven by GitHub's verified organisation badge (nebius)

Agents
Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, OpenCode
License
Apache-2.0 (permissive)
Pin
b1924depinned Oct 9, 2026
Context cost
about 5,600 tokenswhen the skill loads
GitHub stars
7on nebius/nebius-ps-services
Scripts
4 scripts

What it does

“Use when the user asks for one local Git commit without push: inspect the full diff, stage repo-wide, validate, commit, and report. Also handles exact worktree-integration commits; not for PRs.”
The author's description, from SKILL.md

Generated summary from the skill's own text

The commit skill creates one local Git commit with a complete repository-wide diff, validating changes before staging and committing. It accepts user-provided messages or generates one from the diff, and stops on unsafe content, conflicts, or detached HEAD. This skill does not push, create branches, or handle pull requests.

When to use it

  • Committing all current local changes without pushing to remote
  • Staging complete repository changes from any project directory
  • Generating concise commit messages from staged diffs automatically
  • Creating bounded delegated commits for worktree integration tasks

When not to

  • When you need to push changes; use commit-push instead
  • For pull request creation, merge, or review operations
  • As a replacement for sdlc-commit on Agentic SDLC checkpoints

Install

These commands install from the publisher's repository and need no account.

skills CLI

npx skills add nebius/nebius-ps-services --skill commit

Installs from the repository's default branch, which may have changed since the commit we checked.

GitHub CLI 2.90 or later

gh skill install nebius/nebius-ps-services commit@b1924de70353e4e688be5017ec608ec9b89c8ff4

Installs the commit we checked (b1924de).

Copy by hand

Download the folder at b1924de and put it in .claude/skills/commit for Claude Code and in .agents/skills/commit for Codex, Cursor, Gemini CLI, GitHub Copilot and OpenCode.

Skill folder on GitHub at b1924de

Copy for your agent

A prompt for any coding agent. It installs this skill from GitHub at the pinned commit, copies the files as they are, and runs nothing unless you ask.

Prompt

Install the agent skill "commit" by Nebius.

Source, pinned to commit b1924de70353e4e688be5017ec608ec9b89c8ff4:
https://github.com/nebius/nebius-ps-services/tree/b1924de70353e4e688be5017ec608ec9b89c8ff4/skills/commit

1. Download that folder at exactly this commit. Do not use the default branch.
2. Copy every file unchanged, keeping the folder structure, into .claude/skills/commit/ for Claude Code and .agents/skills/commit/ for Codex, Cursor, Gemini CLI, GitHub Copilot and OpenCode. Create the folders if they do not exist.
3. Do not edit the files, and do not run any script in the folder unless I ask you to.
4. When you are done, list the files you copied and where you put them.

Listing and checks: https://www.agentnative.dev/skills/nebius/commit

SKILL.md at the pin

---
name: commit
description: "Use when the user asks for one local Git commit without push: inspect the full diff, stage repo-wide, validate, commit, and report. Also handles exact worktree-integration commits; not for PRs."
---

# Commit

## Help

For `$commit --help` or `$commit -h` (including native Claude forms), return concise help and stop before
any workflow step. State the purpose and invocation policy. Show exact usage
for every public action. Describe each public action, positional
argument, and flag in one concise line, including `-h, --help`; say "No
additional public flags" when there are no others. Use only the documented
public interface. For internal or coordinator-only skills, state that boundary
and that no standalone public workflow action exists. After the selected
`SKILL.md` is loaded, help is report-only: do not call any additional tools,
inspect project state, or modify files, private state, Git, or external systems.
Never expose private helper actions or flags or treat help as workflow
authorization.

## Agent Compatibility

Before workflow reads, load `../global-context-management/references/agent-hosts.md`
for selected-home paths, native identity and required runtime setup.

Use `$commit` in Codex, `/commit` in Claude Code, or
`/skills:commit` in the Claude plugin. Dollar-prefixed skill examples
refer to the same named skill on either host; use the native invocation syntax.
Preserve the declared invocation policy, approvals and workflow ownership.
Use available native tools; an unavailable required capability is a blocker,
never permission to bypass a guard or claim unobserved behavior.

## Purpose

Create one local commit for the current branch as quickly as is safely
reasonable, including a complete monorepo diff that spans any number of
project folders. This skill is for ordinary Git commits, not publishing or
Agentic SDLC checkpoints.

## When To Use

- Committing all current local changes on the current branch without pushing.
- Staging complete repository changes with repo-root `git add -A`, regardless
  of the project, service, chart, app, or package directory where the agent starts.
- Using the user's exact commit message when provided.
- Generating a concise commit message from the staged diff when the user does
  not provide one.
- Reporting a clear no-op when there is nothing to commit.
- Executing one bounded delegated commit for a fresh explicit
  `$worktree integrate`: an eligible ordinary child first or its primary source
  second, at the exact preflight branch and head.

## When Not To Use

- Do not push; use `commit-push` when the user explicitly asks to push.
- Do not create, open, update, review, merge, or close pull requests.
- Do not replace `sdlc-commit`; that skill is only for Agentic SDLC feature
  checkpoints with SDLC evidence and local run state.
- Do not create branches, switch branches, pull, merge, rebase, cherry-pick,
  revert, stash, reset, amend, squash, or repair divergence.
- Do not stage a partial pathspec or project-scoped path. This skill always
  stages the complete repository diff with `git add -A` from the Git root.
- Do not run formatters, dependency updates, tests, or broad repair commands
  just to create the commit unless the user separately requested them.

## Inputs

- The exact Git worktree and current branch selected by the user's session.
- A current nonauthorizing root-turn receipt from the prompt hook, or exact delegated
  owner evidence from Task Implementer or Worktree. The root agent binds the receipt to an authorized
  commit or commit-push task with private `begin`; delegated owners retain
  their existing evidence path.
- An optional user-provided commit message; otherwise the reviewed candidate
  must be coherent enough to summarize truthfully.

## Required Reads

- Repository-root branch, operation, conflict, status, staged, unstaged, and
  untracked state before preparation.
- The hook-provided receipt digest and canonical authorization and claim paths,
  or coordinator-provided delegated paths;
  never discover or substitute an alternate helper or private-state path.
- The complete temporary-index candidate summary and enough focused diff to
  review every changed project, root file, risky path, and generated artifact.
- In delegated mode, the exact current assignment, preflight, branch, head,
  and owner evidence returned by the owning workflow.

## Writes

- Owner-private task, checkpoint, attempt authorization and claim state under the selected agent
  home transaction root; these records contain bounded identities and digests,
  not prompt, message, diff, or repository file content.
- The real Git index and one local commit only inside the owning transaction
  helper after the reviewed candidate is revalidated.
- No push, PR, publication, external-system write, branch operation, or
  Agentic SDLC state mutation.

## Process

1. Enter the repository root.
   - Reject caller-provided repository-shaping Git environment before any Git
     discovery or mutation. The transaction's private preview is the only
     permitted `GIT_INDEX_FILE` override.
   - Use `git rev-parse --show-toplevel`, then run all Git commands from that
     root.
   - Treat the Git root as the commit scope. Never infer a narrower scope from
     the starting directory.
2. Run fast branch and repository safety checks.
   - Stop on detached `HEAD`.
   - Stop if a merge, rebase, cherry-pick, revert, or bisect is in progress.
   - Stop if unresolved conflicts exist.
   - If the repository default branch is known locally and the current branch
     matches it, stop unless the user explicitly asked to commit on the default
     branch with clear, separate consent for that branch. Only then pass the
     private `--allow-default-branch` assertion; commit-push never permits it.
     Do not perform network fetches only to discover the default branch.
3. Inspect current status.
   - Run `git status --short --branch`.
   - Bind the root task with private `begin` before preparing effects, including
     a clean start. If the worktree is clean, close it with `finish completed`
     and report that there is nothing to commit.
   - Inspect staged and unstaged diffs plus every untracked, renamed, deleted,
     credential-like, environment, key, and generated path before staging.
   - Stop before staging obvious secrets, private endpoints, credentials,
     unclear generated artifacts, or a diff too broad or incoherent to
     summarize truthfully.
4. Prepare one reviewed repository transaction.
   - Classify the actual root-user request semantically. Accept natural-language
     commit requests, polite action questions, and native skill mentions at any
     position. For example, "please commit all changes", "can you commit this?",
     and "commit and push using $commit-push" are action requests. The latter
     belongs to commit-push, which owns the separate publication phase.
     Discussion, help, quoted examples, negation, and requests to edit or repair
     this skill do not authorize Git effects. Skill selection alone never
     grants permission. Never demand that the user repeat an already clear
     request or move a skill token to the beginning.
   - The prompt hook records a nonauthorizing current root-turn receipt. Only
     after classifying an authorized action, pass its exact hook-provided
     digest as `--intent-sha256` and the typed `--requested-action commit` (or
     `commit-push` inside that workflow) to canonical `begin`. Keep this original
     digest as the task key for later prepare/finish calls, even after unrelated
     turns update the receipt. These are private helper arguments, not public
     skill flags. The immutable task grant freezes the action, native session,
     repository, worktree and target. Never hand-write or reset private state.
     Read `references/task-lifecycle.md` for the exact private interface.
     Subagent and generated turns do not supply root receipts. An absent
     `agent_type` remains compatible with primary UserPromptSubmit events.
   - Project lifecycle state is advisory and is never a commit prerequisite.
     Direct preparation relies on the original authorized task grant,
     already-effective repository instructions, and the transaction's own Git,
     branch, Worktree, secret, and workflow-conflict checks. Project lifecycle
     attestations are not required for the current or any sibling scope.
   - A Task Implementer worker does not use root-user intent. Its exact
     one-direct-child authorization is minted by `task-start` from the immutable
     assignment, running task plane, worker session, branch, and base `HEAD`;
     the helper revalidates that evidence before preparation, execution, and
     interrupted-commit recovery.
   - Use the canonical installed `commit_transaction.py prepare` helper with
     the hook-provided authorization and claim paths, exact Git root, and
     current session. Task Implementer returns the same two paths from its
     owner transition inside a transient canonical `commit_context`. A worker
     runs that context's exact `prepare_argv` from its `scope_cwd` and uses
     the returned raw `session_id` from the native host adapter (`CODEX_THREAD_ID` on Codex;
     hook-bound `SKILLS_SESSION_ID` on Claude); it must never use the
     `worker_session_fingerprint_sha256` as `--session-id`. Never invoke a
     source-tree or alternate helper.
   - The helper copies the current real index into a private temporary index,
     runs repo-root `git add -A` there, validates the candidate, and returns the
     exact `git write-tree` candidate plus a one-shot private claim token.
     Preparation must leave the real index and worktree unchanged.
   - Inspect the complete candidate with read-only Git tree/diff commands.
     Stop before execution if it contains unsafe, incoherent, or unexplained
     content. No project lifecycle attestation is required.
5. Execute the exact reviewed transaction.
   - Use the canonical installed `commit_transaction.py execute` helper with
     the same root, session, claim and token, the exact reviewed candidate tree,
     and the final commit message. Run one uncomposed command.
   - For delegated workers, keep the exact Python executable and helper from
     `commit_context`. The hook admits its own interpreter or an exact
     PATH-canonical `python3`/`python3.N`; arbitrary same-name paths and
     wrappers remain denied.
   - Under the selected installation's common-repository lock, the helper revalidates ref, `HEAD`, real
     index, complete porcelain status, candidate tree, and Worktree ownership.
     Any drift makes the claim stale before real staging.
   - The helper alone runs repo-root `git add -A` with no pathspec, verifies the
     staged tree, and runs `git diff --cached --check`.
   - On candidate drift or a failed hook with no commit, preserve the checkout.
     Correct only already-authorized problems, prepare a new candidate, review
     it again and execute its new token under the original task. Do not request
     another commit phrase merely because the previous attempt became stale.
     Semantic repairs outside this task still require their own authorization.
   - Run `git diff --stat HEAD <candidate-tree>` and
     `git diff --name-status HEAD <candidate-tree>` while reviewing.
   - If the candidate tree equals `HEAD^{tree}`, report that there is nothing
     to commit.
   - Inspect candidate filenames and enough focused candidate diff to avoid an
     obviously wrong or unsafe commit message, especially for new config,
     credential-like, environment, key, or generated files.
   - In delegated Worktree mode, continue to pass its reviewed tree plus the
     exact preflight head to Worktree's private integration-commit action; do
     not create a second direct claim.
6. Commit inside the owning transaction.
   - Use the user's exact commit message if provided.
   - Otherwise generate one concise imperative subject line from the staged
     diff. If the diff is too broad or unclear to summarize truthfully, stop and
     ask for a commit message.
   - For an ordinary direct `$commit`, the transaction helper runs
     `git commit -m "<message>"` with normal hooks enabled. Do not run a raw
     `git commit` or raw `git add` outside the helper.
   - For a delegated Worktree commit, let the private Worktree helper create a
     durable source-scoped preparation claim, rerun `git add -A`, compare the
     staged tree to the reviewed tree, and then run the normal-hook commit.
     The claim blocks nested leases, competing integrations, removal, and
     source publication until candidate reservation or explicit abort.
7. Verify, reconcile, and report.
   - Run `git status --short --branch`.
   - In delegated worktree mode, require the same branch, exactly one
     direct-descendant commit from the preflight head, and a completely clean
     checkout. Return that exact commit SHA to the worktree workflow.
   - Compare `HEAD^{tree}` with the reviewed staged tree. If a hook changed the
     committed tree, preserve the commit, mark the claim `REVIEW_REQUIRED`, and
     inspect the complete actual direct-child commit. After that review, use the
     helper's private `review` transition with the exact observed commit and
     tree to complete the claim. Never amend or reset automatically. If the
     actual commit cannot be fully reviewed or is not the clean exact direct
     child, stop with `REVIEW_REQUIRED` intact.
   - A failed hook that did not create a commit and pre-commit candidate drift
     make that attempt `STALE`. Retry safely under the same task after the cause
     is resolved and unchanged base history is proved. Each attempt requires a
     fresh candidate review and token; it does not consume the one-commit limit.
   - An actual commit consumes the standalone task's allowance, including a
     hook-modified result awaiting review. Never create a second standalone
     commit to repair that result. Close a completed or zero-commit task with
     private `finish`; cancellation invalidates unused claims while preserving
     uncertain effects and allowing exact metadata-only result review.
   - Exact crash recovery accepts only the same branch and either the unchanged
     base or one direct-child commit with the reviewed tree. A fresh authorized
     user request may rebind an otherwise unchanged prepared claim; it never uses
     a TTL, process identity, or guessed ownership.
   - Report the branch name, commit hash and message when a commit was created,
     the validation performed, and whether the worktree is clean.

## Calls From Other Skills

A calling skill may use this shared transaction when the root user's authorized
task includes the commit. Interpret the parent task semantically; do not demand
another standalone `$commit` phrase. Skill invocation alone is not authority.
Use the existing delegated Worktree or Task Implementer route for their commits.
For an unmanaged `create-pr` task (including authorized complete publish-*,
review-pr repair and merge-pr review delegation), read
`../create-pr/references/commit-continuation.md`: its private grant permits
successive reviewed commits and corrected no-commit retries until that PR task
finishes. The parent PR skill owns repairs, checks and pushes. Ordinary direct
commit and commit-push tasks permit one actual commit plus safe no-commit retries; active SDLC still owns its
commits. This exception does not grant arbitrary sibling skills indefinite
commit authority or permit raw Git fallback.

## Recommended Commands

- Repository root: `git rev-parse --show-toplevel`
- Current branch: `git symbolic-ref -q --short HEAD`
- Status: `git status --short --branch`
- In-progress operation checks: inspect Git state paths such as `MERGE_HEAD`,
  `rebase-merge`, `rebase-apply`, `CHERRY_PICK_HEAD`, `REVERT_HEAD`, and
  `BISECT_LOG` under the Git directory from `git rev-parse --git-path <name>`.
- Conflict check: `git diff --name-only --diff-filter=U`
- Default branch, local only:
  `git symbolic-ref -q --short refs/remotes/origin/HEAD | sed 's#^origin/##'`
- Candidate preparation: canonical installed commit transaction helper with
  the hook-provided root-turn receipt digest, authorization and private claim paths
- Staging inside the helper: `git add -A` from the repository root, with no
  pathspec
- Staged validation inside the helper: `git diff --cached --check`
- Candidate summary: `git diff --stat HEAD <candidate-tree>`
- Candidate filenames: `git diff --name-status HEAD <candidate-tree>`
- Reviewed staged tree: exact `candidate_tree` returned by preparation
- Commit inside the helper: `git commit -m "<message>"`
- Committed tree: `git rev-parse HEAD^{tree}`
- Commit hash: `git rev-parse --short HEAD`

## Idempotency

- One immutable task grant owns independently reviewed attempt claims; each
  execution requires
  the claim's exact session, token, reviewed tree, repository identity, and
  owner evidence.
- Repeated execution returns the already-proven commit instead of creating a
  second one. Fresh preparation may adopt only an unchanged exact staged state
  or one exact direct child after all applicable ownership checks pass.
- Terminal claims are archived by exact content digest before a later fresh
  transaction replaces the active claim; no timeout or process identity is
  treated as completion evidence.

## Failure Handling

- Candidate drift at an unchanged base permits a fresh reviewed attempt under
  the same task. Scope/owner drift, unexplained history and uncertain effects
  require reconciliation; they do not become retry authority.
- Structured failures return `code`, `retryable`, and `next_action`. Follow that
  disposition rather than asking for a new prompt for every failed attempt.
- A normal hook failure with no commit becomes `STALE`; preserve the real
  index and worktree for diagnosis rather than resetting or unstaging them.
- A created direct child whose tree or checkout is not the exact reviewed
  clean result becomes `REVIEW_REQUIRED`; complete it only through the private
  review transition after inspecting the actual commit and tree.
- Malformed private state, conflicting workflow ownership, default-branch
  ambiguity, merge history, or unverifiable recovery fails closed without
  repository cleanup or a second commit.

## Must Not

- Require an authorized root-user action and its exact current receipt at
  `begin`. Interpret intent semantically; neither a receipt nor a skill mention
  alone grants permission. The task contains no prompt or commit-message text.
  Closed tasks cannot reopen; a fresh authorized task needs its own receipt.
  Root task/authorization/claim schemas never import old root authority.
  Delegated Task Implementer and Worktree protocols stay unchanged.
- Do not use the helper to bypass already-effective repository instructions,
  unresolved Git state, active Worktree ownership, or workflow-owned commit
  policy. Project lifecycle status remains advisory in every case.
- Treat a fresh explicit `$worktree integrate <name>` as permission for only
  the eligible child/source commits that its read-only preflight orders. Do not
  infer delegated permission from a coordinator handoff, active reservation,
  restart, candidate, or ordinary implicit skill selection.
- Treat a Task Implementer worker authorization as permission only in its exact
  assigned worktree/session while the task plane is still running at the
  assigned base. It never authorizes primary-source dirt or another worker.
- Never push, create a PR, dispatch CI, publish artifacts, or call external
  write APIs from this skill.
- Never run `git add -A` outside the repository root.
- Never broadly allow raw Git mutation. Only the exact digest-pinned helper and
  current transaction evidence are admissible.
- Never pass a pathspec to `git add -A`; this skill is intentionally whole-repo
  because multi-folder repositories often have related changes outside the
  current directory.
- Never use `--no-verify`; commit hooks should run normally.
- Never use `--allow-empty`; an empty staged diff is a no-op.
- Stop before committing obvious secrets, private endpoints, credential files,
  generated files with unclear ownership, unresolved conflicts, or staged
  validation failures.
- Stop on known default branch unless the user explicitly authorized committing
  there.
- Never write Agentic SDLC evidence, permissions files, run state, or
  checkpoints. An active Agentic SDLC run rejects the direct helper; use
  `sdlc-commit` inside the SDLC workflow instead.
- In delegated worktree mode, stop if the observed branch or head differs from
  preflight. Never commit a nested/coordinated child or any dirty checkout after
  an integration reservation exists.
- When Task Implementer integration finds dirty primary-source state, require a
  separate fresh explicit `$commit`, then repeat integration. Never auto-commit
  that dirt.

## Completion Criteria

- Direct mode reports either a truthful no-op or one local exact direct-child
  commit whose tree matches the reviewed candidate and whose final checkout is
  clean.
- Delegated mode returns the exact eligible commit to its owner with unchanged
  branch identity and all owner evidence still current.
- Final output names the branch, commit hash and message when created,
  repo-root `git add -A`, validation performed, final status, and any retained
  `STALE` or `REVIEW_REQUIRED` blocker.
- No push, PR, branch rewrite, project-lifecycle prerequisite, or external write
  occurred.

## Learning Loop

When using this skill, capture durable, reusable, public-safe learnings
in the narrowest appropriate surface only when the task contract allows source edits.
For read-only/report-only work, or when a learning is not public-safe,
evidence-backed, in scope, or free of unverified/vendor-specific claims, do not
edit skill sources; report that it was skipped. Do not capture secrets, private
URLs, customer data, raw logs, or one-off local state.

## Output Contract

Return:

- Whether the run committed or no-op'd.
- The current branch.
- The commit hash and commit message when a commit was created.
- Confirmation that staging used repo-root `git add -A`.
- The lightweight validation performed.
- Final `git status --short --branch` interpretation.
- Any blocker that stopped the workflow.

From nebius/nebius-ps-services at b1924de, shown as plain text under its license: Apache-2.0 (permissive). The license text is in LICENSE.

View SKILL.md on GitHub at b1924de

Files

10 files, 287 KB in total. Each file has a sha256 in the pin's manifest; the first 12 characters are shown.

Skill folder on GitHub at b1924de
All 50 skills from Nebius

More from Nebius

  • agent-development

    This skill guides building the NPA chat agent backend using a grounded-first routing architecture that minimizes token spending.

    Claude Code, Codex, Cursor +4 · about 5.7K tokens · pinned 6 days ago

  • author-npa-workflow

    Helps authors write and validate NPA workflow YAML specifications, which are declarative state machines that orchestrate tools via SkyPilot.

    Claude Code, Codex, Cursor +4 · about 3.3K tokens · pinned 6 days ago

  • cosmos

    This skill enables the agent to work with Cosmos, a world model for synthetic data and video generation.

    Claude Code, Codex, Cursor +4 · about 1.6K tokens · pinned 6 days ago

  • cosmos3-npa-workflow

    This skill helps agents write, validate, and submit npa.workflow YAML specifications for Cosmos 3 image and video generation tasks.

    Claude Code, Codex, Cursor +4 · about 1.7K tokens · pinned 6 days ago

  • cosmos3-post-training

    This skill guides planning and understanding of Cosmos3 supervised fine-tuning workflows for robotic policy training on NPA infrastructure.

    Claude Code, Codex, Cursor +4 · about 2.5K tokens · pinned 6 days ago

  • curobo

    This skill enables agents to run NVIDIA cuRobo V2 motion planning for robotic arm pose planning and benchmarking on CUDA GPUs.

    Claude Code, Codex, Cursor +4 · about 1.5K tokens · pinned 6 days ago

All Coding workflow skills
  • planning

    This skill orchestrates multi-workflow projects with dependencies and shared data tables by creating a task graph in the orchestrator via the create-tasks tool.

    by n8nOfficial publisher

    Claude Code, Codex, Cursor +3 · about 2.4K tokens · pinned 6 days ago

  • progressive-building

    This skill implements incremental workflow development for agents building or extending workflows.

    by n8nOfficial publisher

    Claude Code, Codex, Cursor +3 · about 1.4K tokens · pinned 6 days ago

  • workflow-builder

    Builds and edits n8n workflows using the TypeScript SDK.

    by n8nOfficial publisher

    Claude Code, Codex, Cursor +3 · about 14.2K tokens · pinned 6 days ago

  • remotion-upgrade

    This skill upgrades Remotion and its related packages to the latest stable version.

    by RemotionOfficial publisher

    Claude Code, Codex, Cursor +3 · about 460 tokens · pinned 6 days ago

  • remotion-upgrade

    This skill upgrades Remotion and related packages across a project, automatically detecting the package manager and workspaces.

    by RemotionOfficial publisher

    Claude Code, Codex, Cursor +3 · about 460 tokens · pinned 6 days ago

  • final-release-review

    Audits Python SDK release candidates or plans against the previous release, comparing diffs for regressions, API changes, and version compatibility.

    by OpenAIOfficial publisher

    Contains scripts

    Claude Code, Codex, Cursor +3 · about 5.4K tokens · pinned 6 days ago

Open roles at Nebius

The skills listed here are third-party software, collected from public GitHub repositories as a catalog of recent skills and provided as is. Agent Native does not write or control them and is not responsible for any harm, loss or damage caused by installing or using them. Labels and checks describe what we saw at one commit and are not a guarantee. Read a skill's files before you run it. Terms of use