메뉴
Initialize context-harness in a new repository by creating AGENTS.md, CONTEXT.md, NOW.md, and optional PLAN.md.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Route lightweight project context work to the right companion skill: context-init for new repositories, context-catch-up for session resume, and set-goal for long-running goal/loop prompts, and context-maintain for updates, lessons, plan stress-tests, closeout, reflection, and automatic Dream/Compact consolidation, and context-upgrade for upgrades/migrations.
Catch up a fresh agent session or true resumed session from context-harness files before planning or editing; do not use for ordinary mid-session follow-up turns after context is already loaded.
Maintain context-harness during active work: update context, capture lessons, stress-test plans, keep PLAN.md/NOW.md current, and reflect after corrections or failed attempts, with automatic Dream/Compact consideration after maintenance events.
Plan, implement, validate, and deploy context-harness source upgrades, local fleet refreshes, installed skill deployments, or explicit current-v3 layout repairs. Use when the user asks to upgrade context-harness itself, repair stale or partial context files, package lessons from an upgrade, or verify deployment targets.
Turn the current conversation and repository state into a directly usable long-running goal for Codex goal mode, loop mode, or a fresh agent run, including purpose, context, constraints, milestones, verification, stop conditions, and closeout instructions.
| name | context-init |
| description | Initialize context-harness in a new repository by creating AGENTS.md, CONTEXT.md, NOW.md, and optional PLAN.md. |
| user-invocable | true |
| allowed-tools | Read, Write, Edit, Bash, Glob, Grep |
Use this only when a repository has no context-harness layout and should get a
fresh AGENTS.md, CONTEXT.md, NOW.md, and optional PLAN.md.
If a repository already has partial, stale, or custom context-harness files, use
explicit context-upgrade instead. Init does not repair existing layouts.
Create the smallest durable context layer that lets future agents catch up:
AGENTS.md as the automatic Context ContractCONTEXT.md for project facts, operating constraints, language, relationships, and lessonsNOW.md for the current focusPLAN.md only when there is an active multi-step taskGenerated AGENTS.md and CONTEXT.md must include
<!-- context-harness:schema v3 --> so future skill versions can detect the
layout.
Run the generator:
node scripts/context-gen.js
If running from an installed companion skill directory, use:
node ../scripts/context-gen.js
Present the generated ## Suggested Operating Constraints and ## Suggested Workflow as a starting point.
Ask the user to keep only project-specific constraints a future agent would not infer from code or docs, and confirm workflow verification commands.
Write CONTEXT.md, NOW.md, and AGENTS.md from the templates in the root
context-harness skill.
Install context-harness runtime scripts into the target repo:
node <context-harness-skill-dir>/scripts/install-project.js
Run node scripts/context-index.js update so AGENTS.md has a generated
high-level index and .context-harness/ has retrieval cards.
Ask whether there is a multi-step task. If yes, create PLAN.md.
PLAN.md Done Criteria and CONTEXT.md Workflow Verification instead of
durable project-wide objectives.context-upgrade.AGENTS.md small: contract plus generated CONTEXT.md index and a
pointer to hydrate.AGENTS.md can refer to
node scripts/context-index.js update and hydrate.