Make the CLI pretty ✨ (#2793)
* feat: add ⚙ �� prefix to console output for visual distinction
Add a prefix and separator to all console log messages so Cog's
output is visually distinguishable from Docker build output and
model output:
- Info/Debug: dim ⚙ › prefix
- Warn: bold yellow ⚠ › prefix
- Error/Fatal: bold red ⅹ › prefix
Blank lines at info/debug level print without prefix for clean
spacing. Warn/error blank lines keep their prefix to maintain
visual continuity in multi-line error messages.
Output() (stdout, for model predictions) remains unstyled.
* feat: add console.Bold() and highlight dynamic values in messages
Add a Bold() helper to the console package that applies bold
formatting when color is enabled. Use it to highlight key dynamic
values (image names, paths, URLs, hostnames) in user-facing
messages so they stand out within the dim-prefixed lines.
Also remove single quotes around values where bold makes them
visually distinct without needing quoting.
* chore: demote noisy build messages to debug level
Move 'Generating model schema...', 'Using local coglet wheel...',
and 'Using local cog wheel...' from info to debug level. These are
implementation details that clutter normal output but remain visible
with --debug.
* feat: add spacing between build/prediction output phases
Add blank lines after 'Building Docker image...' (before Docker
build output) and after 'Running prediction/training...' (before
model output) to visually separate Cog's framework messages from
external output.
* feat: add console.Success() with green ✓ prefix via Style system
Introduce a Style type that controls the icon/color of a log line
independently from the log level. This allows Success to display at
info level (same filtering) but with a green ✓ prefix.
New styles can be added by extending the Style enum without touching
the level hierarchy.
Apply Success to: image built, cog init completion, login success,
and image pushed messages. Drop the ✅ emoji from init since the
green ✓ prefix replaces it.
* feat: wrap long lines to terminal width with prefix on continuation lines
When stderr is a TTY, wrap console messages to the terminal width so
they don't overflow. Each continuation line gets the same prefix as the
first line for visual continuity.
Wrapping is ANSI-aware: escape codes are treated as zero-width so
colored/bold text doesn't cause premature breaks. Lines break on word
boundaries where possible, with hard breaks as a fallback.
* feat: add InfoUnformatted for prefix-free interactive output
Add InfoUnformatted/InfoUnformattedf methods that write to stderr at
info level without any icon prefix. These are used for conversational
and interactive output (login prompts, instructions) where the gear
prefix would be visual noise.
Long lines are wrapped to terminal width when stderr is a TTY, same
as prefixed output.
Apply to login flows in both Replicate and generic providers: replace
console.Info with console.InfoUnformatted for all user-facing prompts
and instructions. Rename 'CLI auth token:' prompt to 'Token:' and
improve spacing around the prompt.
* fix: improve error message wording and add spacing after run message
- 'Failed to predict' → 'Failed to run prediction' in predict.go
- 'The inputs you passed to cog predict' → 'The inputs you passed' in
predictor.go (shorter, less redundant since the user knows they ran cog)
- Add blank line after 'Running...' message in run.go for visual
separation from Docker output
* refactor: improve init command output style
- Remove leading \n from 'Setting up...' message (prefix handles
spacing), use separate console.Info('') for the blank line after
- Show 'Creating {filename}' (info, before write) instead of
'Created {fullpath}' (success, after write) — less noisy, uses
just the filename since the user knows their working directory
* refactor: move build message and spacing to CLI layer
The 'Building Docker image...' message and surrounding blank lines are
user-facing concerns that belong in the CLI commands, not deep in
pkg/image/build.go. This lets each command control what it shows:
- cog build / cog push: show image name (user-chosen, meaningful)
- cog predict / cog train / cog run / cog serve: hide image name
(ephemeral internal detail the user can't use)
Each CLI command now prints the message + blank line before calling
resolver.Build(), and the pre-ImageBuild blank lines are removed from
image/build.go.
* fix: update integration tests for console output changes
- build_cog_init: match 'Image built as' without 'cog-' prefix since
the success prefix (✔) is now part of the line
- wheel_resolution: use --debug flag since wheel resolution messages
were demoted to debug level
