Module suggest 

ski suggest — turn the telemetry log into concrete, copy-pasteable tuning actions. ski history shows the recall misses and false positives; this closes the loop by saying what to do about them:

• Recall misses (the model self-loaded a skill ski stayed silent on, repeatedly): suggest force = ["<skill>"] when one of the skill’s existing keywords already appears in the missed prompts (so force would have fired as-is), and/or suggest new keywords: for its SKILL.md mined from the recurring content tokens of the missed prompts.
• Repeat false positives (ski injected it across several sessions and the model never once used it): suggest deny = ["<skill>"] — the config key the README calls the most-reached-for one.

Everything here is suggestion, never mutation: ski does not edit the user’s config or SKILL.md files. Analysis is pure (analyze) and unit-testable without IO; only run touches the filesystem. Read-only over the same JSONL ski history reads, and equally tolerant of malformed lines.

Structs

Deny

A skill ski keeps injecting that the model never uses.

Miss

A skill the model keeps finding on its own while ski stays silent, with the action(s) that would close the gap.

Suggestions

Functions

analyze

Analyze a telemetry log against the (optional) index. The index supplies each skill’s existing keywords/description so suggestions don’t propose what is already there and can tell whether force is ready to fire; without it (no index built yet) keyword mining still works, just unfiltered.

run

ski suggest: analyze the telemetry log for host and print actions.