context-kernel: скилл для автоматизации рабочих процессов

--- name: context-kernel description: "Shared operating rules for the OpenClaw 100k Posts system. Defines the workspace contract, what files each skill can read or write, and how artifacts move from research to analysis to concept generation to delivery." metadata: openclaw: always: true user-invocable: false --- # Context Kernel This skill is the shared system layer for this repo. It exists to keep the other skills aligned around one runtime contract: - where shared files live - which skill owns which files - what can be overwritten vs appended - how to pass context between skills without inventing missing state ## Shared Workspace All persistent state lives under `~/clawd/workspace/`. ### Brand Context - `~/clawd/workspace/brand/voice-profile.md` - `~/clawd/workspace/brand/notes.md` ### Research + Analysis - `~/clawd/workspace/content-radar/config.yaml` - `~/clawd/workspace/content-radar/briefings/` - `~/clawd/workspace/content-dna/outputs/` - `~/clawd/workspace/content-dna-batch/dna-cache/` - `~/clawd/workspace/content-dna-batch/reports/` ### Production + Delivery - `~/clawd/workspace/the-forge/outputs/` - `~/clawd/workspace/monday-drop/outputs/` ## Ownership Rules - `voice-lab` owns `voice-profile.md`. - `voice-lab` may append useful discoveries to `notes.md`. - `content-radar` owns briefing files in `content-radar/briefings/`. - `content-dna` owns single-video analysis files. - `content-dna-batch` owns pattern reports and DNA cache. - `the-forge` owns forged concept output. - `monday-drop` owns delivery logs and exported run summaries. ## Read Rules - `voice-lab` may read any file in `workspace/brand/`. - `content-radar` should only read `config.yaml`. - `content-dna` should read `voice-profile.md` only when adapting concepts. - `content-dna-batch` should read Content DNA outputs and its own cache. - `the-forge` should read `voice-profile.md`, pattern reports, and optionally the latest content-radar briefing. - `monday-drop` should read config, reports, and forged outputs. It should not invent missing brand state. ## Write Rules - Never overwrite `voice-profile.md` without showing the user what will change. - `notes.md` is append-only. - Dated run artifacts should create new files instead of overwriting old ones. - If a script or skill cannot produce a promised artifact, say that clearly in the output instead of faking completeness. ## Output Rules - Prefer clean markdown. - Saved output paths should be printed clearly. - If a claim is based on live fetched data, say so. - If a section still needs agent follow-up, mark it explicitly rather than presenting it as finished. ## Fallback Rules - If no `voice-profile.md` exists, skills may fall back to the demo voice where supported. - If Content Radar has no completed briefing, The Forge may still run from a pattern report alone. - If Monday Drop has no delivery credentials, it should still write output files and exit cleanly with delivery skipped.