The orchestrator's fleet-memory discipline: answer ANY question about code that exists across the ~95-repo fleet — "have we built X", "which repo does Y", "what's the exact signature", "what breaks if I touch this", "how is repo Z shaped" — by asking Almanac FIRST (grounded index, repo:path:line receipts on every hit) instead of grep/agent fan-outs, and by trusting or distrusting what comes back correctly (staleness, fallback notes, known blind spots). Includes the voice contract for spoken (Siri) turns: verdict aloud, receipts to the log, never speak a file:line. Use on every fleet-code question, before any cross-repo grep, before writing code that calls a fleet API, and for all voice-mode fleet answers.
Install
npx skillscat add repairman29/chump/almanac-mastery Install via the SkillsCat registry.
/almanac-mastery — Claude Code adapter
Per `.claude/README.md`: this file is a thin adapter. The
capability is the Almanac CLI/MCP; the protocol is
`docs/ALMANAC.md`. Read that doc now and follow
it — tool selection by question shape, the seven trust rules, latency
shaping, and the voice contract all live there, not here.
Routing
- MCP-wired sessions: use the
almanacMCP tools (almanac_search_fleet,almanac_search,almanac_api,almanac_impact,almanac_architecture,almanac_neighbors,almanac_status). - Otherwise shell out to the harness-neutral CLI:
~/Projects/almanac/target/release/almanac --help$ARGUMENTS, if present, is the fleet question itself — answer it per the
protocol doc (Almanac first, receipts always, almanac_status before
load-bearing answers).