| name | llm-ide-handoff-packager |
| description | Convert PDRs, repo-router dispatch boards, project queues, audit findings, and agent tasks into execution-ready handoff packages for Cursor, Codex, Claude, Gemini, GitHub Issues, and local LLM/IDE agents. Use when operationalizing plans across other LLMs or IDEs, creating implementation briefs, generating target-specific task files, or packaging repo work into portable handoffs with allowed actions, blocked actions, acceptance criteria, and stop conditions. |
| skill_api_version | 1 |
LLM / IDE Handoff Packager
Bridge planning and execution by turning governed work packets into target-specific, self-contained handoffs that can be executed without relying on hidden chat context.
Operating rule
Every handoff must be independently executable. Include enough context, constraints, and evidence that the target agent can begin and stop safely without needing the originating conversation.
Core procedure
- Identify the source material. Confirm the input: PDR summary, dispatch board, queue item, or audit finding. Note any missing fields that would block execution.
- Identify the target environment. Ask or infer: Cursor, Codex, Claude, Gemini, GitHub Issue, local LLM, or generic agent.
- Apply the universal handoff contract (see below) to every task regardless of target.
- Adapt for the target environment:
- Cursor — include repo path assumptions, exact files to inspect, edit scope, test/validation commands, stop conditions when files are missing or architecture conflicts appear
- Codex — include implementation goal, explicit diff boundaries, test command, PR summary draft, failure handling
- Claude — include longer project background, decision history, review questions, critique and alternatives request
- Gemini — include broad review target, comparison dimensions, visual/multimodal references
- GitHub Issue — include concise title, labels, linked source docs, acceptance checklist, owner/agent role
- Local LLM — keep compact; use classification/routing tasks; prefer JSONL-friendly outputs
- Validate before output. Confirm each handoff has: a stop condition, named allowed and blocked actions, output expectations, and enough source references for the target to begin.
- Bundle batch handoffs when multiple tasks share a common source; produce one package index plus individual task files.
Universal handoff contract
Every handoff must include:
task_id: string
title: string
source_project: string
source_files: list
objective: string
context: string
allowed_actions: list
blocked_actions: list
inputs_required: list
outputs_required: list
acceptance_criteria: list
stop_conditions: list
review_gate: string
handoff_target: cursor|codex|claude|gemini|github-issue|local-llm|generic
priority: p0|p1|p2|p3
status: ready|blocked|needs-human-decision
Output format
## Handoff Package — [Task ID]
**Target:** [Cursor / Codex / Claude / Gemini / GitHub Issue / Local LLM]
**Priority:** [P0–P3]
**Status:** [ready / blocked / needs-human-decision]
### Objective
...
### Context
...
### Source files
- ...
### Allowed actions
- ...
### Blocked actions
- ...
### Acceptance criteria
- [ ] ...
### Stop conditions
- Stop if: ...
### Review gate
...
Guardrails
- Do not authorize destructive file or repo operations unless explicitly approved in the source task.
- Do not assume hidden context will be available to the target agent; include all needed references inline.
- Insert a
blocked status with a missing-input note instead of guessing at incomplete inputs.
- Keep each handoff bounded and independently executable.
Related references
references/ide-handoff-modes.md
references/output-contracts.md