#8455: feat: add thinking/model config to skills.entries

## Summary Adds `thinking` and `model` fields to `skills.entries.<skill>` config, allowing per-skill overrides for thinking level and model selection (similar to how cron jobs work). ## Changes - **`src/config/types.skills.ts`** — Added `thinking` and `model` to `SkillConfig` type - **`src/config/zod-schema.ts`** — Added validation for new fields in skills.entries schema - **`src/agents/skills/types.ts`** — Added `thinking` and `model` to `SkillCommandSpec` - **`src/agents/skills/workspace.ts`** — Lookup skill config and pass thinking/model to command spec - **`src/auto-reply/reply/get-reply-inline-actions.ts`** — Apply thinking/model as directives when skill command invoked ## Example Config ```json { "skills": { "entries": { "radarr": { "enabled": true, "thinking": "medium", "model": "anthropic/claude-sonnet-4-5" } } } } ``` When a user runs `/radarr <command>`, it uses `medium` thinking and Claude Sonnet instead of session defaults. ## Testing - `pnpm build` ✅ - `pnpm format` ✅ - Manual verification of config loading and directive application Closes #8441 <!-- greptile_comment --> <h2>Greptile Overview</h2> <h3>Greptile Summary</h3> This PR adds per-skill `thinking` and `model` overrides under `skills.entries.<skill>` by extending the config types + Zod schema, plumbing those fields into `SkillCommandSpec` during workspace command spec generation, and applying them as inline directives when a skill command is invoked. The approach fits the existing directive pipeline by reusing the same `InlineDirectives` fields that are normally populated via chat directives (e.g. `/model`, `/thinking`). It also broadens directive parsing to accept both `/directive` and `.directive` prefixes. Notable risk areas are precedence between user-provided directives vs skill-level defaults, and the widened regex matching (`[/.]...`) which can increase false positives in normal prose. <h3>Confidence Score: 3/5</h3> - This PR is likely safe to merge but includes behavior changes that could surprise users in chat directive parsing and precedence. - Core changes are small and well-contained (types/schema + passing fields through), but two behavior shifts stand out: skill config overrides currently take precedence over explicit user directives during skill invocation, and directive regexes now match dot-prefixed tokens which can cause accidental directive activation/stripping in normal text. These are user-facing behavior changes rather than build-breaking issues. - src/auto-reply/reply/get-reply-inline-actions.ts, src/auto-reply/model.ts, src/auto-reply/reply/directives.ts, src/auto-reply/reply/exec/directive.ts <!-- greptile_other_comments_section --> <!-- /greptile_comment -->