- add transition scoring engine to learn sequential workflows per
directory using command skeletons
- prioritize current active git branch over older branches in
suggestions
- restore strict chronological order for history navigation and bypass
AI re-ranking
- add in-memory session history for instant access to just-run commands
- fix PTY prompt not syncing when using arrow keys in the history menu
- fix history tie-breaker to properly prioritize newer commands by
assigning larger IDs
- fix potential mutex deadlock between bufferMu and pty write during
history navigation
- fix workspace git detection to remove depth limits and correctly
identify repositories with detached HEADs
- fix SQLite connection leaks in frecency queries
- fix global state leaks in history tests by snapshotting and restoring
registry states
- modernize string prefix checks using strings.CutPrefix
Enhance lookup mechanics and priority propagation (`spec/lookup.go`)
Remove the requirement of typing a leading `-` when searching for
options, allowing flag suggestions to surface naturally during input
Accurately forward the `Priority` field from registry specs into the
`Suggestion` struct so the scoring engine evaluates relevance properly
Integrate scoring into the main suggestion pipeline
(`root/suggestions.go`)
Collect context signals using `scoring.CollectSignals(cwd, query,
rootCmd, store)` with the live working directory retrieved from
`spec.GetCWD()`
Pass the entire deduplicated list across spec, history and AI items
through `scoring.Score(deduped, signals)` to compute scores and sort by
descending order before rendering on screen
Record execution history in Frecency Store (`root/wrapper.go`)
Save the last executed command to `lastSubmittedCommand` whenever Enter
is pressed or a suggestion is selected
Trigger `store.Record(...)` asynchronously within the `IRIS_CMD_STOP`
hook using a dedicated goroutine guarded by panic recovery and timeout
limits, preventing blocks on the main shell thread
Optimize workspace identification for AI
(`internal/ai/context_provider.go`)
Use `workspace.DetectCached(cwd)` to instantly fetch ecosystem context
across Git, Node, Go, Rust, Python and Docker alongside active git
branch state
This caching mechanism avoids redundant subprocess execution when
outside git repositories, significantly speeding up prompt context
generation
This project is starting to grow, and I've seen flaws in the initial
project structure design
I think I'll restructure the entire project directory for better
scalability and also cleaner
## Feat
* feat(ai): add AI suggestion engine, context gathering, and ghost text
overlay (`v0.3.0`)
## Security & Perf
* fix(ai): restrict background `--help` execution to a hardcoded command
allowlist to prevent RCE
* fix(root): defer context cancellation in goroutine to prevent resource
leaks
* perf(ai): implement LRU eviction and maximum size limit for
`ProviderCache`
* perf(ai): truncate command context output
(`CommandContextProvider.Gather`) to 1000 characters to save tokens
## Bug Fixes
* fix(ai): add mutex synchronization and thread-safe snapshotting to
`AIEngine.RegisterProvider` and `GatherDynamicContext`
* fix(ai): return a copy of fresh `Suggestion` in `AIEngine.Suggest` to
prevent cache mutation
* fix(ai): use length-prefixed encoding in `EnvSnapshot.Hash` to prevent
delimiter collisions
* fix(ai): skip variable assignment lines containing `=` when extracting
Makefile targets
* fix(runner): check `scanner.Err()` and return `nil` on scan errors in
justfile generator
* test(ci): rename `git commit` to `git checkout` in overlay test to
resolve CI `typos` false positive
- I think it would be better if one for commands only, one for the spec
handling core logic so people won't be confused by the core logic inside
the commands folder (cleaner I guess)
- Regen docs