feat(repl): bring REPL to feature parity with ZSH shell plugin by tusharmath · Pull Request #2984 · tailcallhq/forgecode

tusharmath · 2026-04-13T10:10:59Z

Summary

Bring the REPL to feature parity with the ZSH shell plugin by adding fzf-based file and command pickers, a syntax highlighter, a starship-style prompt redesign, and a unified forge list file command that both the REPL and the shell plugin now share.

Context

The ZSH shell plugin has long had rich completion (fzf + fd, bat preview, colon-prefix commands, highlighted input) that the built-in REPL lacked. Users switching between shell mode and REPL mode experienced a jarring inconsistency. This PR closes that gap by porting the ZSH experience into the REPL itself, while also sharing the underlying file-listing logic through a new forge list file CLI command so the shell plugin no longer needs a separate fd dependency.

Changes

fzf file picker in REPL — replaces the nucleo fuzzy-matcher with ForgeWidget (fzf). File completions now open an interactive picker with bat-powered syntax-highlighted previews (falls back to cat), matching the shell plugin's @-mention completion experience.
fzf command picker in REPL — replaced the inline suggestion list for / and : commands with an fzf popup that shows each command name and description side-by-side, making command discovery much easier.
Colon-prefix command support — :command now works identically to /command in both completion and execution, matching shell plugin behaviour.
Syntax highlighter — new highlighter.rs that colours slash/colon commands, shell (!cmd) commands, and file mentions ([path]) in real-time as the user types, giving immediate visual feedback.
Starship-style prompt redesign — the left prompt now shows the git branch (with branch symbol); the right prompt shows agent symbol, model name (provider prefix stripped), token count (humanised, e.g. 1.5k), cost, and dims all labels when the session is inactive (no tokens consumed yet).
forge list file CLI command — new forge list --file / forge list files subcommand that walks the workspace with the same rules as fd --type f --type d --hidden --exclude .git and prints one path per line. The shell plugin's completion.zsh now delegates to this command instead of calling fd directly, removing the external dependency.
Walker parallelism and hidden-file support — Walker::get_blocking migrated to WalkBuilder::build_parallel() for better performance on large workspaces. New hidden flag controls dotfile visibility (false = include hidden, matching fd --hidden). .git directory is always excluded.

Key Implementation Details

Walker::max_all() (used by the REPL picker and forge list file) sets hidden: false, so dotfiles are included just like fd --hidden --exclude .git. Walker::default() keeps hidden: true for agents that should not see dotfiles. The parallel walker uses Arc<Mutex<…>> shared state so all threads contribute to the same collection and global limits. The fzf command picker builds CommandRow display objects (name padded to 30 chars + description) and pre-populates the fzf query with whatever the user has already typed after the sentinel.

Use Cases

# Tab-complete a file mention — opens fzf picker with bat preview
> Tell me about @src/<TAB>

# Tab-complete a command — opens fzf picker showing name + description
> /<TAB>   or   :<TAB>

# Real-time highlighting while typing
> /ask explain this [src/main.rs]   ← command is green, file is cyan

# List workspace files from any script or shell alias
$ forge list file --porcelain | fzf

Testing

# Run all crate tests (snapshot tests auto-accepted)
cargo insta test --accept

# Manual REPL smoke-test
cargo run -- repl
# 1. Type `@` then Tab → fzf file picker should open with preview
# 2. Type `/` then Tab → fzf command picker should open
# 3. Type `:` then Tab → same fzf command picker, colon sentinel preserved
# 4. Observe syntax highlighting as you type a slash command + file mention

# Verify the new CLI command
cargo run -- list file
cargo run -- list file --porcelain

Links

Related PR (original fzf file picker): feat(repl): bring REPL to feature parity with ZSH shell plugin #2984

…ver marketing language

…ker and add preview support

…ing to it

…ommands, and file mentions

…lid command error message

…nstants

…dispatch

… from zsh plugin

…ands

tusharmath · 2026-04-13T13:47:54Z

+        };
+        let command = normalised_command.as_str();
+
        // TODO: Can leverage Clap to parse commands and provide correct error messages


Can leverage Clap to parse commands and provide correct error messages - drop the sentinel char and then send it to Clap for parsing

…to clap

…mmand

…directly

… in prompt

The forge REPL strips both `/` and `:` prefixes before Clap (part of the tailcallhq#2984 parity work), which meant the speed-dial slot hook and the `SpeedDialMenu` subcommand inherited the same `/ == :` equivalence when the feature landed. That gave `:1..:9`, `:sd`, and `:speed-dial` two entry points with identical behaviour — one inside the forge TUI and one at the zsh prompt — and muddied the mental model the zsh plugin already owned. Narrow the Rust parser so speed-dial activation is slash-only inside the REPL. Other colon commands (`:compact`, `:sage`, ...) keep the parity they had, because only the speed-dial tokens are gated. - model.rs: before the pre-Clap slot hook, if `command_prefix == ':'` and `bare` is a single digit `1..9`, `speed-dial`, or `sd`, return `AppCommand::Message(input)` so the line falls through as plain chat. `is_slot_digit` is hoisted so the later `/N` branch can reuse it. - tests: new `test_parse_speed_dial_colon_prefix_falls_through_as_message` covers all nine `:N`, `:2 explain this`, `:speed-dial`, `:sd`, and `:sd 3 --clear`. Existing `/N` tests unchanged. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

… REPL" This reverts e995a28. Restoring `:1..:9`, `:speed-dial`, and `:sd` as REPL-usable aliases. The earlier commit special-cased speed-dial by making it slash-only inside the forge TUI while leaving every other command's `/ == :` parity intact. On reflection that asymmetry is the surprise, not the parity — the tailcallhq#2984 design explicitly maps zsh muscle-memory onto the REPL, and speed-dial should not be the one feature that breaks it. Having `:3` mean the same thing inside forge as it does at the shell prompt is the simpler mental model and matches user expectation. The zsh plugin continues to own `:` at the shell level; nothing about that ownership required forge to reject the same tokens inside its own input box. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

The forge REPL strips both `/` and `:` prefixes before Clap (part of the tailcallhq#2984 parity work), which meant the speed-dial slot hook and the `SpeedDialMenu` subcommand inherited the same `/ == :` equivalence when the feature landed. That gave `:1..:9`, `:sd`, and `:speed-dial` two entry points with identical behaviour — one inside the forge TUI and one at the zsh prompt — and muddied the mental model the zsh plugin already owned. Narrow the Rust parser so speed-dial activation is slash-only inside the REPL. Other colon commands (`:compact`, `:sage`, ...) keep the parity they had, because only the speed-dial tokens are gated. - model.rs: before the pre-Clap slot hook, if `command_prefix == ':'` and `bare` is a single digit `1..9`, `speed-dial`, or `sd`, return `AppCommand::Message(input)` so the line falls through as plain chat. `is_slot_digit` is hoisted so the later `/N` branch can reuse it. - tests: new `test_parse_speed_dial_colon_prefix_falls_through_as_message` covers all nine `:N`, `:2 explain this`, `:speed-dial`, `:sd`, and `:sd 3 --clear`. Existing `/N` tests unchanged. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

… REPL" This reverts e995a28. Restoring `:1..:9`, `:speed-dial`, and `:sd` as REPL-usable aliases. The earlier commit special-cased speed-dial by making it slash-only inside the forge TUI while leaving every other command's `/ == :` parity intact. On reflection that asymmetry is the surprise, not the parity — the tailcallhq#2984 design explicitly maps zsh muscle-memory onto the REPL, and speed-dial should not be the one feature that breaks it. Having `:3` mean the same thing inside forge as it does at the shell prompt is the simpler mental model and matches user expectation. The zsh plugin continues to own `:` at the shell level; nothing about that ownership required forge to reject the same tokens inside its own input box. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

The forge REPL strips both `/` and `:` prefixes before Clap (part of the tailcallhq#2984 parity work), which meant the speed-dial slot hook and the `SpeedDialMenu` subcommand inherited the same `/ == :` equivalence when the feature landed. That gave `:1..:9`, `:sd`, and `:speed-dial` two entry points with identical behaviour — one inside the forge TUI and one at the zsh prompt — and muddied the mental model the zsh plugin already owned. Narrow the Rust parser so speed-dial activation is slash-only inside the REPL. Other colon commands (`:compact`, `:sage`, ...) keep the parity they had, because only the speed-dial tokens are gated. - model.rs: before the pre-Clap slot hook, if `command_prefix == ':'` and `bare` is a single digit `1..9`, `speed-dial`, or `sd`, return `AppCommand::Message(input)` so the line falls through as plain chat. `is_slot_digit` is hoisted so the later `/N` branch can reuse it. - tests: new `test_parse_speed_dial_colon_prefix_falls_through_as_message` covers all nine `:N`, `:2 explain this`, `:speed-dial`, `:sd`, and `:sd 3 --clear`. Existing `/N` tests unchanged. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

… REPL" This reverts e995a28. Restoring `:1..:9`, `:speed-dial`, and `:sd` as REPL-usable aliases. The earlier commit special-cased speed-dial by making it slash-only inside the forge TUI while leaving every other command's `/ == :` parity intact. On reflection that asymmetry is the surprise, not the parity — the tailcallhq#2984 design explicitly maps zsh muscle-memory onto the REPL, and speed-dial should not be the one feature that breaks it. Having `:3` mean the same thing inside forge as it does at the shell prompt is the simpler mental model and matches user expectation. The zsh plugin continues to own `:` at the shell level; nothing about that ownership required forge to reject the same tokens inside its own input box. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

The forge REPL strips both `/` and `:` prefixes before Clap (part of the tailcallhq#2984 parity work), which meant the speed-dial slot hook and the `SpeedDialMenu` subcommand inherited the same `/ == :` equivalence when the feature landed. That gave `:1..:9`, `:sd`, and `:speed-dial` two entry points with identical behaviour — one inside the forge TUI and one at the zsh prompt — and muddied the mental model the zsh plugin already owned. Narrow the Rust parser so speed-dial activation is slash-only inside the REPL. Other colon commands (`:compact`, `:sage`, ...) keep the parity they had, because only the speed-dial tokens are gated. - model.rs: before the pre-Clap slot hook, if `command_prefix == ':'` and `bare` is a single digit `1..9`, `speed-dial`, or `sd`, return `AppCommand::Message(input)` so the line falls through as plain chat. `is_slot_digit` is hoisted so the later `/N` branch can reuse it. - tests: new `test_parse_speed_dial_colon_prefix_falls_through_as_message` covers all nine `:N`, `:2 explain this`, `:speed-dial`, `:sd`, and `:sd 3 --clear`. Existing `/N` tests unchanged. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

… REPL" This reverts e995a28. Restoring `:1..:9`, `:speed-dial`, and `:sd` as REPL-usable aliases. The earlier commit special-cased speed-dial by making it slash-only inside the forge TUI while leaving every other command's `/ == :` parity intact. On reflection that asymmetry is the surprise, not the parity — the tailcallhq#2984 design explicitly maps zsh muscle-memory onto the REPL, and speed-dial should not be the one feature that breaks it. Having `:3` mean the same thing inside forge as it does at the shell prompt is the simpler mental model and matches user expectation. The zsh plugin continues to own `:` at the shell level; nothing about that ownership required forge to reject the same tokens inside its own input box. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

tusharmath added 2 commits April 13, 2026 15:12

docs(skills): refine release notes tone guidelines to favor clarity o…

b5bbae8

…ver marketing language

feat(completer): replace nucleo fuzzy matcher with fzf-based file pic…

07d265e

…ker and add preview support

tusharmath changed the title ~~fzf repl tagging~~ feat(completer): replace nucleo fuzzy matcher with fzf file picker and preview support in REPL mode Apr 13, 2026

feat(cli): add list file command and migrate shell plugin file list…

d8ff325

…ing to it

github-actions Bot added the type: feature Brand new functionality, features, pages, workflows, endpoints, etc. label Apr 13, 2026

tusharmath added 3 commits April 13, 2026 16:22

feat(completer): replace command completion list with fzf-based picker

67af95b

feat(highlighter): add syntax highlighter for slash commands, shell c…

393624a

…ommands, and file mentions

feat(repl): add colon prefix support and redesign starship-themed prompt

ef20df4

tusharmath force-pushed the fzf-repl-tagging branch from d61c37c to ef20df4 Compare April 13, 2026 11:20

autofix-ci Bot and others added 3 commits April 13, 2026 11:22

[autofix.ci] apply automated fixes

c31c1a4

fix(prompt): dim model label when prompt is inactive and improve inva…

ec6d800

…lid command error message

[autofix.ci] apply automated fixes

fa8bb7a

tusharmath changed the title ~~feat(completer): replace nucleo fuzzy matcher with fzf file picker and preview support in REPL mode~~ feat(repl): bring REPL to feature parity with ZSH shell plugin Apr 13, 2026

tusharmath and others added 8 commits April 13, 2026 17:15

Merge branch 'main' into fzf-repl-tagging

7b4a40a

chore(config): remove unused fd command detection

477061b

fix(prompt): replace hardcoded ANSI color indices with named color co…

4a23e70

…nstants

refactor(repl): migrate slash commands from json file to enum-driven …

471d289

…dispatch

refactor(shell-plugin): remove doctor and keyboard-shortcuts commands…

23a33a0

… from zsh plugin

[autofix.ci] apply automated fixes

c645dec

[autofix.ci] apply automated fixes (attempt 2/3)

47aee9a

fix(highlighter): apply mention highlighting to text after slash comm…

c5cf49e

…ands

tusharmath commented Apr 13, 2026

View reviewed changes

tusharmath added 7 commits April 13, 2026 19:30

refactor(prompt): remove starship references from comments

12fd97b

refactor(repl): rename SlashCommand to AppCommand and migrate parser …

96779a2

…to clap

fix(repl): remove colon prefix from fzf tag buffer pre-fill

c7938c7

fix(repl): output commit result directly instead of setting buffer co…

e5f367c

…mmand

feat(repl): add optional conversation id arg to switch conversations …

996ef46

…directly

fix(repl): change conversation id arg from vec to option

b71bd9d

fix(repl): show accumulated session cost instead of last request cost…

b251e36

… in prompt

tusharmath force-pushed the fzf-repl-tagging branch from 2c64016 to b251e36 Compare April 13, 2026 15:20

tusharmath mentioned this pull request Apr 13, 2026

[Feature]: Call external editor in REPL #2986

Closed

2 tasks

tusharmath and others added 3 commits April 13, 2026 20:51

Merge branch 'main' into fzf-repl-tagging

b2eee90

[autofix.ci] apply automated fixes

30398aa

[autofix.ci] apply automated fixes (attempt 2/3)

40b218c

tusharmath merged commit 16f54bb into main Apr 13, 2026
8 checks passed

tusharmath deleted the fzf-repl-tagging branch April 13, 2026 17:01

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(repl): bring REPL to feature parity with ZSH shell plugin#2984

feat(repl): bring REPL to feature parity with ZSH shell plugin#2984
tusharmath merged 27 commits intomainfrom
fzf-repl-tagging

tusharmath commented Apr 13, 2026 •

edited

Loading

Uh oh!

tusharmath Apr 13, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

tusharmath commented Apr 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Context

Changes

Key Implementation Details

Use Cases

Testing

Links

Uh oh!

tusharmath Apr 13, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

tusharmath commented Apr 13, 2026 •

edited

Loading