fix(codebuddy): clarify reload and source stats

[hxy91819]

Summary

• clarify that CodeBuddy snapshots hook settings at session startup and needs a restart or /hooks confirmation after external settings edits
• switch CodeBuddy install output to the structured success format with reload and verification guidance
• make stats --by-source easier to inspect by sorting sources by raw observed cost, showing raw/reduced/saved metrics, and explaining historical source unknown entries

Real environment validation

• In an isolated clean environment, tokenjuice install codebuddy --local wrote ~/.codebuddy/settings.json with a managed hooks.PreToolUse entry using matcher Bash.
• A newly started shell/CodeBuddy process with the same HOME could read the installed hook via codebuddy config get hooks.
• A different CODEBUDDY_CONFIG_DIR did not see the hook, and tokenjuice doctor codebuddy correctly reported it disabled for that config dir.
• The full interactive /hooks panel was not fully exercised in the clean TUI run because CodeBuddy stopped at the workspace trust prompt, but the config-level behavior matched the suspected reload/snapshot issue.
• The CodeBuddy hook rewrite path produces tokenjuice wrap --source codebuddy -- ..., and tokenjuice stats --source codebuddy --by-source counts those entries separately.
• On the current real environment, CodeBuddy stats are present as source codebuddy: 361 entries, 428,977 raw chars, 341,033 saved chars.
• Local artifact sampling showed source unknown was historical metadata without a source field: 3,064 missing source values, 0 invalid source values, concentrated from 2026-04-21 through 2026-04-26.

Validation

• pnpm typecheck
• pnpm build
• env -u TOKENJUICE_NO_OMISSION pnpm verify

Notes

The observed CodeBuddy issue appears to be runtime reload behavior rather than install-time settings corruption: CodeBuddy can read the written config, but an already-running session keeps its startup snapshot until restarted or confirmed through /hooks.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.