feat(cli): add /reload to rescan on-disk extensions

[emal-avala]

Summary

New /reload slash command that re-reads skills, subagent definitions, hooks, and MCP servers from disk, prints a one-line summary, and invalidates the cached system prompt so the next turn picks up any changes.

> /reload
Reloaded: 27 skill(s) · 3 agent(s) · 0 hook(s) · 2 MCP server(s)
System prompt cache cleared; changes take effect on next turn.

Why

Previously, after editing a skill / agent file on disk the user had to restart the REPL to see the change reflected in /skills or in a /<skill> invocation. For a tight authoring loop that's unacceptable — the whole point of session persistence is to keep developing without re-establishing context.

QueryEngine already re-reads skills / rules from disk during build_system_prompt, but the prompt itself is cached keyed on a hash of its inputs. /reload clears that cache so the next turn rebuilds unconditionally.

Implementation

• New slash command in crates/cli/src/commands/mod.rs
• QueryEngine::reset_system_prompt_cache() — public helper for any future commands that mutate prompt inputs
• Counts are computed by re-reading the same files the system-prompt builder and coordinator read at startup

Scoped to the extension surfaces agent-code actually supports. A companion bump to include rules can follow once #186 lands.

Test plan

• cargo fmt --all — clean
• cargo clippy --workspace --all-targets -- -D warnings — clean
• cargo test -p agent-code --test smoke — 3/3 pass
• cargo test -p agent-code-lib --lib — 615/616 pass (pre-existing sandbox test failure, unrelated)
• Manual: ./target/debug/agent --help | grep reload shows the command

[chatgpt-codex-connector]

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