Repository navigation
Description
1. Fourteen tips in settings.enhancements.json (183 tips) state things that are not true of v8.761.0.
| Kind | Tip | What v8 has |
|---|---|---|
| Wrong command | "Use bun Inference.ts fast|standard|smart for AI inference" | Inference.ts takes --level <low|medium|high|max> (LIFEOS/TOOLS/Inference.ts:31); the old names fail |
| Missing settings keys | "The counts section in settings.json tracks skills, signals, files, and sessions." | hooks/handlers/UpdateCounts.ts:14: "The settings.json.counts field is dead." |
| "v5.1: contextFiles is empty. …" | no contextFiles key |
|
| "principal in settings.json stores your name, timezone, and voice clone." | no voice-clone field: the principal mirror carries name, pronunciation and timezone (hooks/lib/identity.ts:321-326) |
|
| Retired Algorithm terms | "The Quality Gate checks count, category diversity, testability, anti-criteria, and capabilities." | no gate in LIFEOS/ALGORITHM/ checks count, category diversity or capabilities; the ISA format dropped count floors, and another shipped tip says so |
| "User-story validation runs through the Interceptor skill." / "Gate 4 (browser-based QA validation) runs via the Interceptor skill." | no user-story validation or Gate 4 in the Algorithm or the Interceptor skill | |
| Four "capability selection" tips: "Capability selection consults the skill listing…", "Selecting a capability creates a binding commitment…", "Verification checks every ISC criterion AND confirms every selected capability…", "The capability audit's ISC Improvement step…" | no capability selection or capability audit in LIFEOS/ALGORITHM/ (two agent files, agents/Opus.md and agents/Astra.md, still mention "the capability selection") |
|
| Stale numbers | "The Fabric skill runs 235 prompt patterns…" | skills/Fabric/Patterns has 274 pattern directories with a system.md |
| "Cortex v8.3.0 (the memory system)…" | ARCHITECTURE_SUMMARY.md: Cortex (Memory) v8.4.0 |
|
| "/review-pr runs comprehensive PR review using 7 specialized analysis agents." | not part of the payload; the pr-review-toolkit plugin, version 1.0.1, ships 6 agents |
Two more tips are true but age badly: "52 public skills, 212 intents. Type / to see them all." (neither number matches the payload), and the history note on the Cato and Forge agents folding into Astra.
2. spinner-tips cannot cover the shipped tips. IntegrityCheck.ts checkSpinnerTips reads the tips from USER/CONFIG/settings.user.json and skips the whole check when that file has none ("public install"). The scaffolded overlay is {}. So the tips that are actually live, which the installer writes into settings.json as the installer-owned spinnerTipsOverride (MergeSettings.ts INSTALLER_OWNED_KEYS), are never checked until something puts tips into the overlay.
3. The count rule misreads scoped counts. The rule at IntegrityCheck.ts:1978 matches every "N intents|procedures|workflows" phrase and compares N with the total across all skills (:1966). A tip like "BeCreative has 7 intents" is therefore flagged although the payload has exactly 7 BeCreative intent files.
Suggested direction (yours to pick).
- Drop numbers, versions and dates from tips, so they cannot go stale.
- Point
spinner-tipsat the tips that are live: the installer-owned key, or both sources. - Have it check what a tip names, so that slash names, file names and settings keys resolve against the install.
- Scope count claims to the skill they name, or leave them out.
Activity