jiangxinke/harness-rl/stackplanner-long-task

Claude Code and agent skill by jiangxinke/harness-rl

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Use for long-horizon StackPlanner research, report, or implementation work that needs staged delegation, artifact lineage, failure recovery, and human review. The workflow is perception and scope, outline, optional approval, evidence or implementation, reflection and replan on failure, versioned deliverable, human feedback, revision, verification, then finish.

Install

npx vskill@latest install jiangxinke/harness-rl/stackplanner-long-task

SKILL.md

The same SKILL.md was published earlier as jiangxinke/harness-rl/stackplanner-long-task.

StackPlanner Long Task

Scope

Use this Skill only inside a StackPlanner specialist delegation. The CentralAgent owns control flow; complete only the delegated stage and never delegate again or ask the user directly.

Workflow Profile

Use the smallest profile that satisfies the delegated contract:

  • Direct: no staged workflow for a simple explanation or already-resolved answer.
  • Bounded execution: one precise specialist task followed by verification.
  • Deliberate: ambiguous, high-impact, multi-hop, or substantial deliverable work. Use the stage contract below.

Do not recreate a completed stage. Inspect context_refs.artifact_bodies, context_refs.mandatory_requirements, and the supplied artifact IDs first.

Deliberate Stage Contract

  1. Perception and scope: normalize the objective, deliverable, audience,

constraints, acceptance criteria, assumptions, and only decision-critical unresolved questions. Return at most one batch of 1-5 orthogonal questions.

  1. Outline or plan: define the minimum useful sections or dependency DAG,

required evidence, acceptance criteria, and decisions that need human approval.

  1. Evidence or implementation: execute the bounded task, preserve source URLs or test evidence, and distinguish facts from inference.
  2. Recovery: if blocked, return a compact failure diagnosis and the smallest viable retry scope. Do not hide partial failure.
  3. Deliverable: put the full result in artifact_content or a created output file; keep the summary under 700 characters.
  4. Revision: apply pinned human feedback before older plans or recalled memory, preserve artifact lineage, and identify the prior version in metadata when available.
  5. Verification: check the delegated acceptance criteria, citations or tests, unresolved gaps, and output location before reporting success.

The stage value describes the delegated work: perception, planning, research, implementation, reporting, revision, or verification. For example, a researcher performing a current-information lookup always uses research.

Artifact Rules

  • Reports, outlines, research bodies, generated files, evidence bundles, and multimodal observations are artifacts.
  • Task memory receives only a compact summary and artifact reference.
  • Never overwrite an earlier report version. Produce a new version linked to its parent.
  • Preserve feedback bindings and source references across revisions.

Output

Return one JSON object only:

json
{
"summary": "compact result for the CentralAgent",
"artifact_content": "full artifact body or null when files were created",
"artifact_type": "outline | research_observation | perception_observation | report_revision | generated_file",
"artifact_metadata": {
"created_paths": [],
"sources": [],
"verification": [],
"unresolved_gaps": []
}
}

Security

  • Security scanPassed, 90 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

Repository stars
442
Repository forks
53
7-day trend score
30

Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/jiangxinke/harness-rl/stackplanner-long-task/badge)](https://verified-skill.com/skills/jiangxinke/harness-rl/stackplanner-long-task)

Something wrong with this skill or its page? Tell us.