mission-forge

Converts feature ideas into executable ContextHub missions with validated specs, phases, and agent assignments.

Updated May 12, 2026
One-click install
npx skills add https://github.com/contexthub-md/marketplace --skill mission-forge-contexthub-md
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: mission-forge
Source: https://github.com/contexthub-md/marketplace/tree/main/contexthub/skills/mission-forge
Command: npx skills add https://github.com/contexthub-md/marketplace --skill mission-forge-contexthub-md

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Turning a raw feature or refactor idea into a structured, executable project plan requires research, codebase validation, task breakdown, and agent coordination — doing this manually is slow and error-prone, and hallucinated file paths or invalid references waste entire agent tasks. ## Core Features & Use Cases - End-to-end mission creation: Researches an idea across the project knowledge base, library docs, and web, then validates every code reference against the OCXP code graph before creating the mission. - Two creation modes: Mode 1 lands a hand-authored plan in one call via ch_create_mission_with_plan (execute-ready); Mode 2 lets AgentCore generate the plan for large or fuzzy missions via pipeline_mode="plan" or "process". - Fully-specced tasks: Produces SMART goals, phased todos with ten-field specs, typed references, dependency ordering, DoD gates, agent roles, model routing, and git worktree setup. - Use Case: Say "forge a mission to add rate limiting to the ask API" and receive a complete mission with research docs, validated code references, ordered phases, and per-agent branches ready for execution. ## Quick Start Ask the agent to forge a mission for your feature idea, for example: forge a mission to migrate the docs knowledge base from AOSS to S3 Vectors.

Frequently Asked Questions about mission-forge

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I turn a feature idea into an executable mission plan?

Run mission-forge with your idea, for example "forge a mission to add rate limiting to the ask API". It researches the idea, validates code references against the OCXP code graph, then creates the full mission with goals, phases, specced todos, and agent assignments in one call.

What is the difference between ch_create_mission_with_plan and pipeline_mode plan?

ch_create_mission_with_plan (Mode 1) lands your hand-authored plan execute-ready in one call with no AgentCore involvement. pipeline_mode="plan" (Mode 2) lets AgentCore research the mission and generate the plan, stopping at review for human approval. The two modes are mutually exclusive per mission.

Why did my mission creation return PLAN_ERROR and leave an orphan mission?

The mission row is created before the plan is validated, so a schema error rolls the mission back to status open and leaves a planless orphan. Common causes are using "command" instead of "run" in dod_gate, bare strings in spec.references, or unknown spec keys. Fix the payload, then reuse or delete the orphan.

Does mission-forge work without an indexed project?

No. It requires an active project with indexed repositories; otherwise you get OCXP_UNAVAILABLE or NO_PROJECT and must run /contexthub:connect first. Research still works without OCXP, but references cannot be validated, so forging on an unindexed project is blocked.

What fields are allowed in a task spec?

Exactly ten fields survive: name, location, spec_type, summary, inputs, outputs, behavior, constraints, edge_cases, and references. Any other key such as problem, root_cause, or verification is silently dropped, so fold that content into summary and edge_cases.

When should I use AgentCore to generate the plan instead of authoring it?

Use Mode 2 when the mission is large, fuzzy, or needs deep research before tasks can be named. Invest in a rich description over 100 characters plus goals to clear the input-readiness gate, since AgentCore receives only the title and description as its brief.