feat(gain): add per-project token savings with -p flag

[heAdz0r]

Summary

Adds per-project scoping to rtk gain so users can see token savings for the current project instead of global aggregates.

Problem: Users working on multiple projects see combined statistics in rtk gain. There's no way to answer "how much does rtk save in this project?"

Solution: rtk gain -p filters all statistics to the current working directory (and subdirectories).

Usage

# Global view (default, unchanged)
rtk gain

# Project-scoped view (new)
rtk gain -p
rtk gain --project

# Works with all existing flags
rtk gain -p --daily
rtk gain -p --format json
rtk gain -p --history

Implementation Details

Files Changed

File	Changes	Purpose
src/tracking.rs	+120	project_path column, migration, filtered query methods
src/gain.rs	+80	project scope resolution, scope-aware display
src/main.rs	+15	--project/-p Clap flag

Total: 215 lines added, 50 lines modified

tracking.rs

• Schema migration: Adds project_path TEXT column via ALTER TABLE (idempotent, backward-compatible)
• Auto-recording: current_project_path_string() captures canonical cwd on every command execution
• Index: idx_project_path_timestamp for fast project-scoped queries
• NULL normalization: Pre-existing rows get empty string (migration runs once)
• Filtered API: All query methods get _filtered(project_path: Option<&str>) variants:
  • get_summary_filtered()
  • get_all_days_filtered()
  • get_by_week_filtered()
  • get_by_month_filtered()
  • get_recent_filtered()
• SQL filter: WHERE (?1 IS NULL OR project_path = ?1 OR project_path LIKE ?2) — exact match + subdirectory prefix match
• Original unfiltered methods delegate to _filtered(None) (zero behavior change)

gain.rs

• resolve_project_scope(): Resolves --project flag to canonical cwd path
• shorten_path(): Abbreviates long paths for display (/Users/foo/bar/project → /.../bar/project)
• Scope-aware header: "RTK Token Savings (Project Scope)" with path display
• All exports (JSON, CSV) pass project scope through

Backward Compatibility

• Existing databases: Migration adds column with empty default. All existing data remains accessible.
• Existing CLI: rtk gain without -p behaves identically (filtered with None returns all rows).
• No new dependencies added.

Security Compliance

Critical files check

• src/tracking.rs is a critical file (per security-check.yml)
• No shell execution: All changes are SQLite queries via rusqlite parameterized statements
• No user input in SQL: Project path comes from std::env::current_dir(), not CLI arguments
• No Cargo.toml changes: No new dependencies

Verification

• cargo build — compiles clean
• cargo clippy — no errors
• cargo test tracking — 7 passed, 1 pre-existing failure (unrelated test_default_db_path)
• cargo fmt — formatted

---

🤖 Generated with Claude Code

[pszymkowiak]

LIKE ?2 without a trailing % wildcard is an exact match — subdirectory prefix matching doesn't work. project_path LIKE '/home/user/project/' will never match /home/user/project/src.

Fix: Some(format!("{}{}%", p, std::path::MAIN_SEPARATOR))

But then _ and % in path names become LIKE wildcards (e.g. my_project matches myXproject). Consider using GLOB instead of LIKE, or add an ESCAPE clause.

[pszymkowiak]

The UPDATE commands SET project_path = '' WHERE project_path IS NULL in Tracker::new() runs on every startup, not just once. After the first migration it's a no-op query on every
command invocation. Consider wrapping it in a schema version check or checking if the column already has a default.

[heAdz0r]

@pszymkowiak Both issues have been addressed and the branch has been rebased on latest master (conflicts resolved):

1. LIKE without trailing wildcard — Fixed in commit 722e053. Replaced LIKE with GLOB throughout. project_filter_params() now returns GLOB patterns with * wildcard instead of LIKE with %. This also fixes the _ character issue (GLOB treats _ as literal, LIKE does not).

• Tests added: project_filter_glob_pattern, project_filter_glob_underscore_safe

2. UPDATE on every startup — The UPDATE is now guarded by SELECT EXISTS(SELECT 1 FROM commands WHERE project_path IS NULL). It only runs when there are actual NULL values to migrate. After the first successful migration, the EXISTS check returns false immediately (cheap O(1) check) and the UPDATE is skipped entirely.

Additionally, the ALTER TABLE ADD COLUMN now includes DEFAULT '', so new rows never have NULL project_path.

Branch rebased cleanly on master (merged dashboard viz from #129). Ready for re-review.