Skip to content

feat(ghost-text): command prediction - #166

Open
versenilvis wants to merge 48 commits into
mainfrom
feat/ghost-text-prediction
Open

feat(ghost-text): command prediction#166
versenilvis wants to merge 48 commits into
mainfrom
feat/ghost-text-prediction

Conversation

@versenilvis

@versenilvis versenilvis commented Sep 28, 2026 •
edited
Loading

Copy link
Copy Markdown
Owner

In the next v0.8.0

2026-09-28_23-30-27.mp4

New config

[core]
prediction = true

  • This is an independent ghost text from our main ghost text, its job to predict next command you may wanna use

  • How to use it?, use right arrow to expand it and only right arrow works for this
    It means "Tab" -> choose the current selection in menu, expand the first ghost text, no matter the current mode
    Right arrow -> expand second ghost text (which is the prediction)

  • As you can see in the video above, the first one is tab, the second one is right arrow

image

Key changes

  • Directory Scoping: Ranks suggestions across 5 workspace tiers so commands from other projects never leak into unrelated folders
  • Context Validation: Verifies recipes and scripts against justfile, package.json, Makefile, and disk within a strict 15ms deadline
  • Sequence Prediction: Tracks command transitions in SQLite to suggest the next step based on the previous command
  • Dedicated Keybindings: Routes Tab to menu selection and Right Arrow to ghost text prediction
  • Cross-Shell Testing: Adds end-to-end pseudo-terminal tests across zsh, bash, and fish
  • Fix pre-fix match in fuzzy search of history suggestion

currently still using and working to improve the prediction algorithm

@versenilvis versenilvis mentioned this pull request Sep 29, 2026
Closed
@versenilvis
fix(ui): prioritize prediction on right arrow instead of accepting me…
317da75 
…nu item
@versenilvis
refactor: remove scratch tests and redundant sequence fallback query
02b13aa
@versenilvis
feat(scoring): add workspace root detection and schema migration
9eefa76
@versenilvis
feat(scoring): implement candidate tiering and scope ranking
5c80029
@versenilvis
perf(scoring): optimize global query and make scope count lazy
124181e
@versenilvis
fix(scoring): create database backup only during schema migration
c14c64a
@versenilvis
feat(ctxcheck): context validation engine and ast parser
f36fb6f
@versenilvis
feat(predict): wire context validation and atomic buffer matching
39df9f3
@versenilvis
test(tui): add scope prediction integration tests and ci shell checks
b5b5824
@versenilvis
docs(dev): document prediction architecture and scoping rules
1103691
@versenilvis
docs(dev): update scoring architecture with workspace tiers and seque…
dc75953 
…nce mining
@versenilvis
test: fix rows.Err checks and string concatenation in tui test harnesses
a6d949b
@versenilvis
fix(db): use VACUUM INTO for WAL-consistent backups and secure log pe…
f5d3782 
…rmissions
@versenilvis
fix(db): enforce 0600 VACUUM INTO, abort on failure, and isolate pred…
b3cbb33 
…ict.log
@versenilvis
fix(predict): preserve predict.log across sessions and support test -…
751e900 
…short
@versenilvis
fix(ui): preserve space separator in ghost text suffix
632a0ae
@versenilvis
fix(predict): ignore repeating navigation commands in sequence predic…
5db05ec 
…tion
@versenilvis
fix(input): forward right arrow when no prediction is available
de07247
@versenilvis
fix(input): tab accepts prediction when no explicit menu selection
8e797bf 
when the user has not navigated the menu (userNavigated=false), the
first auto-highlighted item is not treated as selected — tab falls
through to the prediction path instead (mirrors right arrow behavior)
@versenilvis
fix(history): fix prefix search ranking and expand ghost text on tab/…
2f0d07d 
…right arrow
@versenilvis
fix(input): tab accepts menu selection when menu is visible
753ebd7
@versenilvis
docs(scoring): update frecency weights, workspace tiers, and scope co…
b0be3d1 
…unt expression
@versenilvis
fix(ctxcheck): skip option values for make flags and handle attached …
34f27b7 
…directory/file flags
@versenilvis
fix(ctxcheck): classify bun test as free and return unknown for unrec…
d418c82 
…ognized npm bare subcommands
@versenilvis
fix(ctxcheck): only treat fish and/or as connectors at command position
a3eeb30
@versenilvis
fix(ctxcheck): return free for interpreter module and inline-code flags
19696df
@versenilvis
test(scoring): copy history.db to temp dir in real db benchmark
3a48809
@versenilvis
fix(scoring): track bootstrap sequences with bgWg and bounded timeout
c62ad11
@versenilvis
fix(scoring): normalize cwd across writes, lookups, and fallback queries
be03b93
@versenilvis
fix(wrapper): snapshot naiveBuffer under bufferMu in prediction accep…
87eadc7 
…t paths
@versenilvis
fix(wrapper): fallback to menu ghost text on right arrow when no rela…
fff86df 
…ted prediction
@versenilvis
refactor(scoring): consolidate duplicate candidate queries and backfi…
bcd24df 
…ll loop
@versenilvis
refactor(wrapper): deduplicate acceptLine in tab and right arrow paths
c213643
@versenilvis
fix(test): typo
5c199ba
@versenilvis versenilvis changed the title feat: ghost text prediction feat(ghost-text): command prediction Sep 30, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@versenilvis