plan-one-shot

Create Antigravity-style implementation plans with strict review gates, line-item IDs, and no code changes until the user explicitly says Execute Plan.

Safety Notice

This listing is imported from skills.sh public index metadata. Review upstream SKILL.md and repository scripts before running.

Copy this and send it to your AI assistant to learn

Install skill "plan-one-shot" with this command: npx skills add ishk-sftckz/exspecso/ishk-sftckz-exspecso-plan-one-shot

Plan One Shot

Replicate Antigravity plan-mode behavior using three artifacts and strict review gates.

Hard Rules

  1. Before approval: plan only. No code edits, no mutating commands, no commits.
  2. Always produce or update these artifacts together:
    • .exspecso/<plan-name>/task.md
    • .exspecso/<plan-name>/implementation_plan.md
    • .exspecso/<plan-name>/walkthrough.md (stub only during planning)
  3. Match plan depth to scope.
  4. Block execution until user says exactly: Execute Plan.
  5. Do not skip artifact generation even for short tasks.
  6. After approval: execute strictly in task.md order and update statuses continuously.
  7. For existing projects, preserve architecture, naming, folder layout, and dependency choices unless the user asks for a change.
  8. Prefer minimal diff; do not introduce new abstractions/libraries unless justified by at least two concrete reuse points or a clear blocker.
  9. First call in a new conversation defaults to new planning. Do not infer an active plan from .exspecso/ history.
  10. Only resume prior plan artifacts when the user explicitly asks to continue/resume an existing plan.

Output Control Rules

  • Scope-fit depth: small tasks -> short plans; large tasks -> comprehensive plans.
  • No speculative engineering: avoid unrequested future-proofing.
  • No vague placeholders like "update logic"; describe exact changes.
  • Keep verification realistic and runnable in the current repository.
  • Prefer adapting existing modules over creating new ones.
  • Keep tool/process notes out of main workstream bullets unless they directly affect implementation steps.

Change Block Rules

Every file block must answer: what changes, where, and why. Nothing more.

  1. One block per logical change - Group files that move/modify together.
  2. Folder-level when 3+ files - Use folder path, list files inside.
  3. File-level only when unique logic - If a file needs different treatment, separate it.
  4. Implicit imports - Don't list every import update; state once at group level.
  5. One bullet = one action - No compound sentences.

Good:

#### [MOVE] [_features/resume/](file:///abs/path/resume/)
- From `insights/`: agents/resume-parser.ts, utils/resume.ts, service.ts
- Import paths update to `@/_features/resume`

Bad:

#### [CREATE] [_features/resume/agents/resume-parser.ts]
- Move resume parser agent from insights/agents/resume-parser.ts
- Update import from @/_features/ai/models

#### [CREATE] [_features/resume/utils/resume.ts]
- Move extractText and validateFileSize from insights/utils/resume.ts

Required Plan Structure

implementation_plan.md must follow this order:

  1. Title
  2. Objective - clear and concise, scaled to scope
  3. Proposed Changes
  4. Workstreams
  5. File blocks using this format:
    • #### [MODIFY] [path/to/file.ts](file:///absolute/path/to/file.ts)
    • #### [CREATE] [path/to/new-file.ts](file:///absolute/path/to/new-file.ts)
    • #### [MOVE] [path/to/dest-folder/](file:///absolute/path/to/dest-folder/) for reorgs (list files inside)
    • #### [DELETE] [path/to/old-folder/](file:///absolute/path/to/old-folder/) for removals
  6. Verification Plan

Use --- between major workstreams.

Artifact Lifecycle

All artifacts for one plan must live under a single date-prefixed folder:

  • .exspecso/YYYY-MM-DD-<plan-name>/

<plan-name> rules:

  1. Lowercase.
  2. Keep letters, numbers, and spaces before slug conversion.
  3. Convert spaces/underscores to -.
  4. Remove filler words when possible (the, a, an, for, to, of).
  5. Keep first 3-6 meaningful words.
  6. Max length 36.
  7. Collapse repeated - and trim edges.

Conversation Session Gate

Treat plan state as conversation-scoped first, filesystem-scoped second.

  1. At the first skill invocation in a conversation, assume there is no active plan session.
  2. Default decision on first invocation:
    • Decision: Create New Plan
    • Reason: no active plan in this conversation.
  3. Do not scan .exspecso/ to auto-select or continue an older plan on first invocation.
  4. Continue an existing plan only when the user explicitly requests it (for example: "continue existing plan", "resume plan", "continue implementing the plan") and provides a clear reference.
  5. Valid reference examples:
    • folder name like 2026-02-16-my-plan
    • direct path like .exspecso/2026-02-16-my-plan/implementation_plan.md
    • explicit plan title that uniquely maps to one folder
  6. If the user asks to continue but the reference is missing or ambiguous, ask one targeted clarification for the exact plan folder/path before proceeding.
  7. Once a plan is created or resumed in the current conversation, treat it as the active plan session and apply the replan gate for subsequent requests.

Replan Decision Gate

For every new request after an active plan session exists, decide first:

  • Decision: Continue Current Plan
  • Decision: Create New Plan

Provide one short reason.

Create a new plan when objective/success criteria change materially, a new subsystem is introduced, architecture shifts significantly, or planned updates would rewrite roughly more than 40% of the current plan.

If there is no active plan session in the current conversation, do not apply this gate and create a new plan instead.

task.md Template

# <Task Name>

## Planning
- [/] Explore codebase and identify gaps
- [/] Write implementation plan

## Execution
- [ ] <work item>
- [ ] <work item>

## Verification
- [ ] <verification item>

Rules:

  • During planning, keep Planning items as [/].
  • Keep Execution/Verification unchecked until execution starts.
  • Keep items concise and action-oriented.

walkthrough.md Planning Stub

# <Feature Name> Walkthrough

## Status
Planning complete. Execution not started.

## Planned Scope
- <summary bullet>

## Verification Plan
- See `implementation_plan.md`.

Execution Tracking

After receiving Execute Plan:

  • Mark Planning items [x].
  • Mark exactly one in-progress item as [/].
  • Move [/] forward one item at a time.
  • Never leave more than one [/] item.
  • End with all checklist items marked [x].

Response Contract

  • If approval is missing, end with: Plan ready for review. Awaiting plan review comments.
  • If approved, start execution and respond with: Execute Plan received. Starting execution from task.md Execution section.

Source Transparency

This detail page is rendered from real SKILL.md content. Trust labels are metadata-based hints, not a safety guarantee.

Related Skills

Related by shared tags or category signals.

Coding

HTML to Markdown

Convert HTML↔Markdown for web clipping, clean notes, and batch content migration. Use when users ask 网页转Markdown/Markdown转HTML/批量转换. Supports local files, UR...

Registry SourceRecently Updated
Coding

Podfetcher Tools

Search podcasts, browse episodes, and fetch podcast transcripts from Podfetcher using the bundled Node.js CLI, SDK, or MCP server.

Registry SourceRecently Updated
Coding

test

Extract and categorize expenses from receipts or statements, map to GL codes, check compliance with policies, and flag anomalies for review.

Registry SourceRecently Updated
Coding

CEO Master

Transforms the agent into a strategic CEO and orchestrator. Vision, decision-making, resource allocation, team dispatch, scaling playbook from €0 to €1B. Use...

Registry SourceRecently Updated