Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/whichguy-ask-agent
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: agent-building
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
- Runs as
- Python package
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/ask-agent
What's inside
- Skill name
ask-agent- Description
- A delegation skill, not an agent type. Ask native agents to work in the background, continue useful work in the main conversation, and incorporate their results when they return. Use for "ask an agent", named agent roles, parallel delegation, or launch-and-notify work.
- License
- MIT
- Instructions
- 422 lines in SKILL.md · view on GitHub
Before installing
- Ships executable scripts
Files (11)
- SKILL.md
- consumer-owned-workspace.md · references/
- current-workspace.md · references/
- git-integration.md · references/
- host-capabilities.md · references/
- native-lifecycle.md · references/
- result-handoff.md · references/
- task-interface.md · references/
- workspace-operations.md · references/
- ask_agent_task.py · scripts/
- ask_agent_workspace.py · scripts/
SKILL.md (first lines)
# Ask agent Delegate through this session's native agents. The bundled helper owns Git workspace preparation and eligible cleanup; the native host owns execution and return; the parent owns acceptance and integration. `ask-agent` selects this skill, not an agent type. Do not implement another model launcher, SDK client, subprocess harness, scheduler, or file watcher to simulate native delegation. The current workspace is the default for delegation: a fresh worker runs in the caller's own checkout and branch, with no second worktree or repository. A helper-managed worktree and a consumer-owned workspace are explicit selections, never an automatic interpretation of an existing worktree, a speed request, a request for fresh context, or a request to change code. Default to a fresh background worker while the parent continues useful work. An explicit request to wait takes precedence. Each invocation adds to the initiating conversation's pending work; it does not replace earlier jobs. On the default standalone current-workspace route, `mode: edit` changes only the write set the parent declares from the request; the parent verifies the edits, and the worker does not commit. Questions, digests and reviews use `mode: report-only` and make no filesystem edits. A standalone isolated worktree is an explicit choice: `mode: edit` defaults to commit delivery, `delivery_mode: patch` selects a patch handoff, and `mode: ephemeral` permits only scoped scratch writes whose findings return without code integration. A standalone worktree `mode: edit` includes bringing the verified contribution back into the designated caller or integration checkout by default. The parent performs that integration and validates the combined result before reporting the request SUCCEEDED. Current-workspace edits are accepted in place, while `report-only` and `ephemeral` results have no code integration. A consumer-owned route instead has in-place parent acceptance followed by its consumer's separately owned final caller delivery. A successful worker return alone is not completion. If integration or final delivery is blocked, report the blocker, worker outcome and retained locations. An explicit review-only or return-without-integration request overrides the integration default. Follow the selected compact or legacy delivery contract and preserve caller dirty state. ## Select the route before preparing This card declares `ask-agent/consumer-owned-workspace/v1` and `ask-agent/current-workspace/v1`. Select exactly one route before any helper preparation or native dispatch: - **Current workspace** is the standalone default. A fresh worker runs in the caller's own checkout and branch, with no helper worktree, receipt or close. Use compact `mode: edit` with a write set for changes; edits are not committed automatically. Use `mode: report-only` for a question, digest or review. For standalone work, the task interface below performs the checks described in [Current workspace](references/current-workspace.md). - **Standalone worktree** requires explicit selection with `workspace: worktree`. The compact task interface prepares the snapshot and supports `edit`, `report-only`, and findings-only `ephemeral` modes. Worktree edits default to commit delivery; select `delivery_mode: patch` only when patch delivery is intended. See [Task interface](references/task-interface.md). - **Legacy/orchestrated helper-managed** requires `workspace_route: helper-managed` or an invoking consumer's explicit contract. It retains its direct helper workflow and `patch`, `commits`, or `report-only` delivery rules below. Plan Dispatcher, ShipLoop's parallel chain and a standalone improve-agent request select this existing contract; do not silently route them through the compact standalone interface. - **Consumer-owned** requires an invoking consumer to explicitly set `workspace_route: consumer-owned` and `delivery_mode: in-place`, and to supply the complete workspace/ownership contract in [Consumer-owned workspace](references/consumer-owned-workspace.md). Its worker writes in the consumer's already-bound candidate; helper delivery modes do not apply. An ordinary invocation that selects neither consumer-owned nor helper-managed uses the current-workspace default. Plan Dispatcher, ShipLoop and improve-agent never use it. A consumer that explicitly requests consumer-owned but lacks this card capability or any required contract field stays pending with its available locators. Do not silently switch that incomplete consumer request to the helper-managed route, shared writes, an inherited context, or a second worktree. The currently declared consumer is the `improve-agent` skill, composing a ShipLoop whole-skill Improve child. It supplies one fresh executor for the complete bound Improve invocation. A standalone `improve-agent` request selects the helper-managed route instead, and inline `improve` starts no agent. This is a Codex native pilot with evidence from bounded native fixtures. Earlier helper-managed host evidence does not qualify this composition on Codex, Claude, Grok, or another host. ## Standalone task interface For standalone current-workspace and isolated worktree tasks, use [Task interface](references/task-interface.md). Bind `scripts/ask_agent_task.py` from this selected skill package. Supply the task, expected result, source, exact write set, `workspace`, and `mode` to `prepare`. Only a worktree `edit` accepts `delivery_mode`; omitted means `commits`, and explicit `patch` is supported. Current `edit` and `report-only` have no delivery mode. `ephemeral` requires a worktree and a nonempty write set; it returns findings only and has no commit, patch, or integration. Current editing requires `exclusive_writer: true`; worktree preparation requires `writers_quiescent: true`. These are parent attestations, not locks. When parent and worker share access to local files, select `prepare --packet` with a new absolute path outside the repositories, Git metadata and selected package. The script retains the binding and emits a short native assignment; `worker-start` validates its pinned digest and returns the complete actionable task. Use the emitted command arguments, including the expected digest, without reconstructing them. The parent retains the packet locator and hashes. This transport satisfies the standalone task's context and authority requirements below through the validated `action_packet.task.current_learnings` and `action_packet.task.authorization`; the worker reads them before task work. For hosts without shared file access, omit `--packet` and carry the existing complete inline binding. Do not silently switch transport after a failure. Retain the emitted binding or packet metadata and launch the native worker with its emitted assignment. The worker runs `worker-start` and `worker-finish`; after collecting the actual native result, the parent runs `verify-return`. These scripts do not launch agents, integrate contributions, or clean up worktrees. The parent owns semantic acceptance, any commit/patch integration, combined checks, and later cleanup. The compact route does not replace the legacy orchestrated or consumer-owned contracts below. Its source implementation and local evidence do not establish release or universal host support. Successful checks replace the corresponding compact-route manual steps; do not repeat preparation or assemble a second worker brief. Resolve failed checks before dispatch and never
From whichguy/skill-craft, MIT.
More from whichguy/skill-craft
- 0.592score
rubric-eval
whichguy
Evaluate a prompt change against a rubric over a scenario catalog: run arms on a subject model (Grok by default), grade blind with one evidence-first judge…
SkillAFreshness: A★ 1Claude Code +9 - 0.592score
plan-dispatcher
whichguy
Use when executing or resuming an agreed dependency graph with one main dispatcher and parallel native workers: claim ready steps, preserve launch state…
SkillAFreshness: A★ 1Claude Code +9 - 0.592score
adversarial-review
whichguy
Have an independent model try to break a proposed change and return its findings as testable hypotheses, each with the smallest experiment that would refute it.
SkillAFreshness: A★ 1Claude Code +9 - 0.592score
improve
whichguy
Use when a repository candidate needs a deliberate review-and-improvement loop: use recent Git history, make warranted changes, run meaningful checks, and…
SkillAFreshness: A★ 1Claude Code +9 - 0.592score
review-coverage
whichguy
Add a post-ship improve-to-exhaustion directive to a plan, or run that directive after implementation.
SkillAFreshness: A★ 1Claude Code +9 - 0.589score
in-parallel
whichguy
Use for "in parallel" or when one goal should be divided among independent native agents and returned as one verified aggregate.
SkillAFreshness: A★ 1Claude Code +9 - 0.588score
shiploop-e2e-audit
whichguy
Run the ShipLoop test harness and audit its retained graph, review, test, product and incremental-change evidence.
SkillAFreshness: A★ 1Claude Code +9 - 0.588score
backchain
whichguy
Use when an implementation task needs a dependency-aware plan before coding: backward planner / backchain / precondition-first planning, dependency DAGs…
SkillAFreshness: A★ 1Claude Code +9 - 0.587score
architect
whichguy
Design system architecture and make technology decisions.
SkillAFreshness: A★ 1Claude Code +9 - 0.587score
plan-test
whichguy
Generate comprehensive tests for code.
SkillAFreshness: A★ 1Claude Code +9 - 0.587score
improve-agent
whichguy
Use when Improve should run in a fresh native agent instead of this conversation: start one agent that runs /improve in its own context, keep useful parent…
SkillAFreshness: A★ 1Claude Code +9 - 0.587score
improve-system-prompt
whichguy
Benchmark and compare system prompt variants (V2/V2a/V2b/V2c) for Sheets Chat by running test scenarios through the real GAS-side ClaudeConversation pipeline.
SkillAFreshness: A★ 1Claude Code +9
Signals
- Stars
- 1
- Forks
- 0
- Open issues
- 0
- Last commit (this folder)
- Oct 11, 2026 (18 days ago)
- Last release
- Aug 15, 2026 (2 months ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/whichguy/skill-craft
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from whichguy
- 0.592score
plan-dispatcher
whichguy
Use when executing or resuming an agreed dependency graph with one main dispatcher and parallel native workers: claim ready steps, preserve launch state…
SkillAFreshness: A★ 1Claude Code +9 - 0.592score
rubric-eval
whichguy
Evaluate a prompt change against a rubric over a scenario catalog: run arms on a subject model (Grok by default), grade blind with one evidence-first judge…
SkillAFreshness: A★ 1Claude Code +9 - 0.592score
improve
whichguy
Use when a repository candidate needs a deliberate review-and-improvement loop: use recent Git history, make warranted changes, run meaningful checks, and…
SkillAFreshness: A★ 1Claude Code +9 - 0.592score
adversarial-review
whichguy
Have an independent model try to break a proposed change and return its findings as testable hypotheses, each with the smallest experiment that would refute it.
SkillAFreshness: A★ 1Claude Code +9 - 0.592score
review-coverage
whichguy
Add a post-ship improve-to-exhaustion directive to a plan, or run that directive after implementation.
SkillAFreshness: A★ 1Claude Code +9 - 0.589score
in-parallel
whichguy
Use for "in parallel" or when one goal should be divided among independent native agents and returned as one verified aggregate.
SkillAFreshness: A★ 1Claude Code +9
Cite this page
ask-agent skill. Agent Mart. https://agentmart.dev/skills/whichguy-ask-agent (last updated 2026-10-24)