feat: rtk session — Show RTK adoption across Claude Code sessions

[FlorianBruniaux]

Use case

rtk session shows my last 10 Claude Code sessions with the RTK vs raw command ratio, so I can see if my hook is working.

Scope MVP (v1)

Single command, zero flags. Displays the 10 most recent sessions.

Technical approach: reuse provider.rs directly

ClaudeProvider already exposes:

• discover_sessions(project_filter, since_days) -> Vec<PathBuf> — finds JSONL session files
• extract_commands(path) -> Vec<ExtractedCommand> — parses JSONL, returns command + output_len

What session_cmd.rs does:

1. ClaudeProvider.discover_sessions(None, Some(30)) → all sessions last 30 days
2. For each session: extract_commands(path) → Vec<ExtractedCommand>
3. Count: total = cmds.len(), rtk_count = cmds.iter().filter(|c| c.command.starts_with("rtk ")).count()
4. Calculate adoption %, format output_len
5. Sort by mtime desc, take top 10
6. Display table

Zero JSONL parsing duplication — reuses the trait as-is.

Target output

RTK Session Overview (last 10)
------------------------------------------------------------------
Session            Date        Cmds   RTK   Adoption  Output
------------------------------------------------------------------
abc123de           Mar 10      42     38    90% @@@@.   24K
fed987cb           Mar 09      18     12    67% @@@..    8K
------------------------------------------------------------------
Average adoption: 83%
Tip: Run `rtk discover` to find missed RTK opportunities

(ASCII progress bar with simple blocks for terminal compatibility)

Files to create/modify

File	Action
src/session_cmd.rs	Create (~250 lines, shorter thanks to provider reuse)
src/main.rs	Add mod session_cmd, Commands::Session, routing
scripts/test-all.sh	Add smoke test assertion

Reuse existing

• discover::provider::{ClaudeProvider, SessionProvider, ExtractedCommand}
• tracking::estimate_tokens()
• Table pattern from gain.rs
• dirs::home_dir()

Error handling

• ~/.claude/projects/ absent → "No Claude Code sessions found."
• Corrupted JSONL → handled by provider (skip line)
• Session with no Bash commands → skip
• Implicit limit: 10 sessions max (perf guard)

Tests

• Inline tests with JSONL fixtures in tempdir
• Test RTK vs raw count on mock ExtractedCommand
• Test progress bar rendering (0%, 50%, 100%)
• Test empty session → skip

Verification

cargo fmt --all && cargo clippy --all-targets && cargo test
./target/debug/rtk session   # on real setup
hyperfine 'rtk session'      # target <500ms (I/O bound, JSONL parsing)

Out of scope (v2, defer)

• --project filter by current project
• --days N temporal filter
• --format json
• --verbose per-session command detail

Acceptance criteria

• rtk session displays last 10 sessions with adoption %
• Reuses provider.rs without JSONL parsing duplication
• Sessions with no Bash commands are skipped
• cargo test passes, clippy clean
• CHANGELOG.md + README updated