ai-frontier-dispatchlisted

# AI Frontier Dispatch Produce one personalized briefing in two separate halves: frontier and builders; then markets, deals, and careers. Filter for the reader instead of summarizing a generic news feed. ## Load the right guidance 1. Read [references/method.md](references/method.md) for gathering, scoring, evidence, and output rules. 2. In Claude Code, also read [adapters/claude.md](adapters/claude.md). 3. In Codex, also read [adapters/codex.md](adapters/codex.md). 4. Do not load both adapter files unless comparing harness behavior. ## Resolve configuration Look for a user-supplied path, `AI_FRONTIER_DISPATCH_CONFIG`, `~/.ai-frontier-dispatch/config.yaml`, then current task context. Use [config/dispatch.example.yaml](config/dispatch.example.yaml) only as the schema and starter. Before a first personalized run, obtain the reader profile, interests, exclusions, career track, requested window, and GitHub profile when repo scouting is requested. Never write personal configuration into the public repository unless the user explicitly asks. ## Separate ingestion from judgment Fan out independent research streams concurrently. Use a cost-efficient model for extraction when the harness permits model tiering. Keep candidate scoring, contradiction handling, claim classification, and final synthesis on the strongest available model. If tiering or parallel agents are unavailable, preserve the logical separation with batched searches and a distinct final synthesis pass. Never cla