btw_app(): parse error: premature EOF

[mjfrigaard]

I've been using btw to develop a Shiny app-package. The code for this application is stored in a repo with multiple branches

Background

I've configured btw/ellmer using the code below:

if (interactive()) {
  require(ellmer, quietly = TRUE)
}
if (interactive()) {
  require(btw, quietly = TRUE)
}
options(
  btw.client = ellmer::chat_anthropic(
    system_prompt =
    "You are an expert R/Python programmer who loves explaining complex topics to non-technical audiences. 
    - When writing R code, use base R functions, but follow the tidyverse style guide.     
    - Avoid using `for` loops and prefer functional programming patterns like `apply()` or `purrr`.    
    - When creating graphs/plots, use `ggplot2`. 
    - If writing R Shiny code, use `bslib` for all layout functions (unless explicitly  instructed otherwise).
    - If writing Python Shiny code, use shiny core (not express) to build apps and include explanations in comments. ",
    model = "claude-sonnet-4-5-20250929"
    )
)

This allows me to use the btw_client() to create a chat client:

chat <- btw_client()

And I've been able to create the btw.md file (and run btw_task_create_btw_md())

use_btw_md(scope = "project")
btw_task_create_btw_md(mode = "console", client = "anthropic")

After installing the development version of shinychat, I was able to launch btw_app()

btw_app()

All of these details are available here (and outputs are here).

Issue

When I run the btw_app() in the 30_llm-btw branch of sqp, I see the following:

btw_app()

Prompt and response 1

Expand the downloadable R Markdown report to include all the graphs and tables in the Shiny app.

"I'll help you expand the R Markdown report to include all the graphs and tables from the Shiny app. Let me first examine the current report structure and then the modules to understand what visualizations and tables are available."

Project Files - Find report template files

Project Files - List all module files to understand app structure

Project Files - List R files with correct path

Project Files - List files in the R directory

"Now let me read the current report template and examine the key modules:"

Read report_template.Rmd - Read current report template

Read nav_ui.R - Read nav_ui to understand app structure

Read mod_report.R - Read mod_report to understand report generation

"Now let me examine the helper functions and key modules to understand the data structures:"

Read mod_counts_bar.R - Read counts bar module to understand count visualizations

Read mod_counts_waffle.R - Read counts waffle module

Read mod_dist_box.R - Read distribution box module

Read mod_dist_raincloud.R - Read distribution raincloud module

Read mod_awards_point.R - Read awards point module

Read count_bar_plot.R - Read helper functions for plotting

Read name_case.R - Read name_case helper function

"Now let me check the table modules to understand the data structure better:"

Read mod_counts_tbl.R - Read counts table module

Read mod_dist_tbl.R - Read distribution table module

Read mod_awards_tbl.R - Read awards table module

"Perfect! Now I have a complete understanding of the app structure. Let me create a comprehensive R Markdown report that includes all visualizations and tables. I'll create an expanded version that includes:"

1. Compare section (already exists) - scatter plots

2. Counts section - horizontal bar plots, waffle charts, and count tables

3. Distributions section - box plots, raincloud plots, and summary tables

4. Awards section - scatter plots with labels and awards tables

Error 1

An error occurred:

parse error: premature EOF
                                       {"_intent": "Create expanded re
                     (right here) ------^

Prompt and response 2

It seems like you stopped before you finished.

"You're right! Let me continue and finish reading the table modules:"

Read mod_counts_tbl.R - Read counts table module

Read mod_dist_tbl.R - Read distribution table module

Read mod_awards_tbl.R - Read awards table module

"Great! Now I have a complete understanding of the app structure. Let me create an expanded R Markdown report template that includes all graphs and tables from the four main sections: Compare, Counts, Distributions, and Awards."

Error 1

An error occurred:

parse error: premature EOF
                                       {"_intent": "Create expanded re
                     (right here) ------^

Console output

The error in the console is below:

Warning: ERROR: An error occurred in `chat_append_stream(id="chat-chat")`
Caused by error:
! parse error: premature EOF
                                       {"_intent": "Create expanded re
                     (right here) ------^
Warning in append_stream_task$invoke(client, "chat", input$chat_user_input) :
  ERROR: An error occurred when invoking the ExtendedTask.
Caused by error:
! parse error: premature EOF
                                       {"_intent": "Create expanded re
                     (right here) ------^
Warning: ERROR: An error occurred in `chat_append_stream(id="chat-chat")`
Caused by error:
! parse error: premature EOF
                                       {"_intent": "Create expanded re
                     (right here) ------^
Warning in append_stream_task$invoke(client, "chat", input$chat_user_input) :
  ERROR: An error occurred when invoking the ExtendedTask.
Caused by error:
! parse error: premature EOF
                                       {"_intent": "Create expanded re
                     (right here) ------^
Warning: ERROR: An error occurred in `chat_append_stream(id="chat-chat")`
Caused by error:
! parse error: premature EOF
                                       {"_intent": "Create expanded re
                     (right here) ------^
Warning in append_stream_task$invoke(client, "chat", input$chat_user_input) :
  ERROR: An error occurred when invoking the ExtendedTask.
Caused by error:
! parse error: premature EOF
                                       {"_intent": "Create expanded re
                     (right here) ------^

It looks like the _intent argument from one the tools?

Session info

sessioninfo::session_info()

─ Session info ──────────────────────────────────────────────────────────────────
 setting  value
 version  R version 4.5.1 (2025-06-13)
 os       macOS Tahoe 26.1
 system   x86_64, darwin20
 ui       X11
 language (EN)
 collate  en_US.UTF-8
 ctype    en_US.UTF-8
 tz       America/Phoenix
 date     2025-11-19
 pandoc   3.8.2.1 @ /usr/local/bin/ (via rmarkdown)
 quarto   1.8.26 @ /Applications/quarto/bin/quarto

─ Packages ──────────────────────────────────────────────────────────────────────
 ! package     * version    date (UTC) lib source
 P askpass       1.2.1      2024-10-04 [?] CRAN (R 4.5.0)
 P bsicons       0.1.2      2023-11-04 [?] CRAN (R 4.5.0)
 P bslib         0.9.0      2025-01-30 [?] CRAN (R 4.5.0)
   btw         * 0.0.6.9000 2025-10-22 [1] Github (posit-dev/btw@9fc3621)
 P cachem        1.1.0      2024-05-16 [?] CRAN (R 4.5.0)
 P chromote      0.5.1      2025-04-24 [?] RSPM
   cli           3.6.5      2025-04-23 [1] CRAN (R 4.5.0)
   coro          1.1.0      2024-11-05 [1] CRAN (R 4.5.0)
 P credentials   2.0.3      2025-09-12 [?] RSPM
 P curl          7.0.0      2025-08-19 [?] CRAN (R 4.5.1)
   digest        0.6.38     2025-11-12 [1] RSPM (R 4.5.0)
 P dplyr         1.1.4      2023-11-17 [?] CRAN (R 4.5.0)
   ellmer      * 0.4.0      2025-11-15 [1] RSPM (R 4.5.0)
 P evaluate      1.0.5      2025-08-27 [?] CRAN (R 4.5.1)
 P fastmap       1.2.0      2024-05-15 [?] CRAN (R 4.5.0)
 P fontawesome   0.5.3      2024-11-16 [?] CRAN (R 4.5.0)
 P fs            1.6.6      2025-04-12 [?] CRAN (R 4.5.0)
 P generics      0.1.4      2025-05-09 [?] CRAN (R 4.5.0)
 P gert          2.1.5      2025-03-25 [?] CRAN (R 4.5.0)
 P gh            1.5.0      2025-05-26 [?] RSPM (R 4.5.0)
 P gitcreds      0.1.2      2022-09-08 [?] RSPM (R 4.5.0)
   glue          1.8.0      2024-09-30 [1] CRAN (R 4.5.0)
 P htmltools     0.5.8.1    2024-04-04 [?] CRAN (R 4.5.0)
 P httpuv        1.6.16     2025-04-16 [?] CRAN (R 4.5.0)
 P httr2         1.2.1      2025-07-22 [?] CRAN (R 4.5.1)
 P jquerylib     0.1.4      2021-04-26 [?] CRAN (R 4.5.0)
 P jsonlite      2.0.0      2025-03-27 [?] CRAN (R 4.5.0)
 P knitr         1.50       2025-03-16 [?] CRAN (R 4.5.0)
 P later         1.4.4      2025-08-27 [?] CRAN (R 4.5.1)
   lifecycle     1.0.4      2023-11-07 [1] CRAN (R 4.5.0)
 P magrittr      2.0.4      2025-09-12 [?] CRAN (R 4.5.1)
 P memoise       2.0.1      2021-11-26 [?] CRAN (R 4.5.0)
 P mime          0.13       2025-03-17 [?] CRAN (R 4.5.0)
 P openssl       2.3.4      2025-09-30 [?] RSPM
 P otel          0.2.0      2025-08-29 [?] CRAN (R 4.5.1)
 P pillar        1.11.1     2025-09-17 [?] CRAN (R 4.5.1)
 P pkgconfig     2.0.3      2019-09-22 [?] CRAN (R 4.5.0)
 P processx      3.8.6      2025-02-21 [?] CRAN (R 4.5.0)
   promises      1.5.0      2025-11-01 [1] RSPM (R 4.5.0)
 P ps            1.9.1      2025-04-12 [?] CRAN (R 4.5.0)
   R6            2.6.1      2025-02-15 [1] CRAN (R 4.5.0)
 P rappdirs      0.3.3      2021-01-31 [?] CRAN (R 4.5.0)
 P Rcpp          1.1.0      2025-07-02 [?] CRAN (R 4.5.1)
   renv          1.1.5      2025-07-24 [1] RSPM
   rlang         1.1.6      2025-04-11 [1] CRAN (R 4.5.0)
 P rmarkdown     2.30       2025-09-28 [?] RSPM
 P rstudioapi    0.17.1     2024-10-22 [?] CRAN (R 4.5.0)
   S7            0.2.1      2025-11-14 [1] RSPM (R 4.5.0)
   sap         * 0.0.0.1000 2025-11-19 [1] local
 P sass          0.4.10     2025-04-11 [?] CRAN (R 4.5.0)
   sessioninfo   1.2.3      2025-02-05 [1] CRAN (R 4.5.0)
 P shiny       * 1.11.1     2025-07-03 [?] CRAN (R 4.5.1)
   shinychat     0.2.0.9000 2025-11-18 [1] Github (posit-dev/shinychat@8578e67)
 P sys           3.4.3      2024-10-04 [?] CRAN (R 4.5.0)
 P tibble        3.3.0      2025-06-08 [?] CRAN (R 4.5.0)
 P tidyselect    1.2.1      2024-03-11 [?] CRAN (R 4.5.0)
   vctrs         0.6.5      2023-12-01 [1] CRAN (R 4.5.0)
 P websocket     1.4.4      2025-04-10 [?] RSPM
   withr         3.0.2      2024-10-28 [1] CRAN (R 4.5.0)
 P xfun          0.53       2025-08-19 [?] CRAN (R 4.5.1)
 P xtable        1.8-4      2019-04-21 [?] CRAN (R 4.5.0)
 P yaml          2.3.10     2024-07-26 [?] RSPM

 [1] /Users/mjfrigaard/Library/Caches/org.R-project.R/R/renv/library/sap-28a55b39/macos/R-4.5/x86_64-apple-darwin20
 [2] /Users/mjfrigaard/Library/Caches/org.R-project.R/R/renv/sandbox/macos/R-4.5/x86_64-apple-darwin20/c8cd77eb

 * ── Packages attached to the search path.
 P ── Loaded and on-disk path mismatch.

─────────────────────────────────────────────────────────────────────────────────

[gadenbuie]

I don't think it's related to the _intent tool argument, which all btw tools have, because you can see the tool intent in the screenshots of previous tool responses.

Instead, I think there's likely either some kind of character encoding problem or the model just failed at writing out a complicated string (maybe also due to character encoding problems?).

I can recreate the error by ending the JSON stream early

utf8_json <- '{"_intent": "Create expanded report", "name": "Café", "price": 5}'

partial <- substr(utf8_json, 1, 50)

jsonlite::fromJSON(partial)
#> Error: parse error: premature EOF
#>                                        {"_intent": "Create expanded re
#>                      (right here) ------^

Are you asking the model to update an existing file? Are there any usual characters in that file? (Could you point me to the file directly?)

[mjfrigaard]

Ah it might be the yaml header/code chunks in the R Markdown report (found here).

I've noticed quarto/rmarkdown can be problematic with LLMs...

[mjfrigaard]

Hi @gadenbuie --I adjusted the system prompt in my .Rprofile and changed the prompt sent to the btw_app():

# .Rprofile
options(
  btw.client = ellmer::chat_anthropic(
    system_prompt =
    "You are an expert R/Python programmer who loves explaining complex topics to non-technical audiences. 
    - When writing R code, use base R functions, but follow the tidyverse style guide.     
    - Avoid using `for` loops and prefer functional programming patterns like `apply()` or `purrr`.    
    - When creating graphs/plots, use `ggplot2`. 
    - If writing R Shiny code, use `bslib` for all layout functions (unless explicitly  instructed otherwise).
    - If writing Python Shiny code, use shiny core (not express) to build apps and include explanations in comments. 
    - When writing R Markdown reports, return the code chunks un-fenced (i.e. ```r and not ```{r})",
    model = "claude-sonnet-4-5-20250929"
    )
)

The application launches without any issues, and I've changed the prompt to a more specific question:

"What code could I add to the Counts section of inst/rmd/report_template.Rmd to ensure all the graphs and tables from this tab are returned in the downloadable report?

Only return R code (no the R Markdown syntax)."

This model looks through the package files without any errors:

It returns the following code block:

tryCatch({
  # Extract parameters
  start_year <- params$count_values$start_year
  end_year <- params$count_values$end_year
  chr_var <- name_case(as.character(params$count_values$chr_var), case = "lower")
  x_lab <- name_case(as.character(params$count_values$chr_var))
  
  # Filter data by year range
  counts_data <- subset(
    sap::movies,
    thtr_rel_year >= start_year &
    thtr_rel_year <= end_year
  )
  
  # Prepare data with normalized names
  counts_nms <- name_case(names(counts_data))
  counts_data <- setNames(object = counts_data, nm = counts_nms)
  
  # Calculate summary statistics
  movie_count <- length(unique(counts_data$`Imdb Url`))
  time_span <- abs(start_year - end_year)
  
  cat("\n\n### Summary\n\n")
  cat(paste("**Total Movies:**", movie_count, "\n\n"))
  cat(paste("**Time Span:**", time_span, "years\n\n"))
  cat(paste("**Year Range:**", start_year, "-", end_year, "\n\n"))
  cat(paste("**Grouping Variable:**", x_lab, "\n\n"))
  
  # Create horizontal bar plot
  cat("\n\n### Horizontal Bar Chart\n\n")
  print(
    ggplot2::ggplot(
      counts_data,
      ggplot2::aes(
        x = forcats::fct_rev(
          forcats::fct_infreq(
            !!rlang::sym(x_lab)
          )
        )
      )
    ) +
    ggplot2::geom_bar(
      ggplot2::aes(fill = !!rlang::sym(x_lab)),
      show.legend = FALSE
    ) +
    ggplot2::coord_flip() +
    ggplot2::scale_fill_manual(values = sap::clr_pal12) +
    ggplot2::labs(
      x = NULL,
      y = "# of Movies",
      fill = x_lab
    )
  )
  
  # Create waffle plot
  cat("\n\n### Waffle Chart\n\n")
  counts_data$chr_var <- as.character(counts_data[[x_lab]])
  
  movies_iron <- ggwaffle::waffle_iron(
    counts_data,
    ggwaffle::aes_d(group = chr_var)
  )
  
  print(
    ggplot2::ggplot(
      data = movies_iron,
      ggplot2::aes(x = x, y = y, fill = group)
    ) +
    ggwaffle::geom_waffle() +
    ggplot2::scale_fill_manual(values = sap::clr_pal12) +
    ggplot2::labs(
      x = "", y = "",
      fill = x_lab
    ) +
    ggwaffle::theme_waffle()
  )
  
  # Create summary table
  cat("\n\n### Count Summary Table\n\n")
  tbl_data <- counts_data |>
    dplyr::group_by(.data[[x_lab]]) |>
    dplyr::summarise(n = dplyr::n(), .groups = "drop") |>
    dplyr::arrange(dplyr::desc(n))
  
  print(gt::gt(tbl_data))
  
}, error = function(e) {
  cat("\n\n**Error creating counts visualizations:**\n\n", e$message)
})

I am now getting an error after the final comment from the model:

BTW I have shiny::devmode() set to TRUE.

Warning: Error in [: Can't subset rows with `tokens$role == "user"`.
✖ Logical subscript `tokens$role == "user"` must be size 1 or 4, not 0.
shiny devmode - Turning on full stack trace. To disable, call `options(shiny.fullstacktrace = FALSE)`
This message is displayed once every 8 hours.
  105: <Anonymous>
  104: signalCondition
  103: signal_abort
  102: cnd_signal
  101: <Anonymous>
  100: signalCondition
   99: signal_abort
   98: cnd_signal
   97: stop_indicator_size
   96: <Anonymous>
   95: vec_as_location
   94: withCallingHandlers
   93: subclass_row_index_errors
   92: vectbl_as_row_location
   91: vectbl_as_row_index
   90: [.tbl_df
   89: [
   88: observe
   87: <observer:observeEvent(chat_tokens())>
   86: ..stacktraceon..
   85: valueFunc
   84: ..stacktraceon..
   83: contextFunc
   82: env$runWith
   81: withCallingHandlers
   80: domain$wrapSync
   79: promises::with_promise_domain
   78: captureStackTraces
   77: force
   76: domain$wrapSync
   75: promises::with_promise_domain
   74: withReactiveDomain
   73: domain$wrapSync
   72: promises::with_promise_domain
   71: ctx$run
   70: ..stacktraceoff..
   69: isolate
   68: func
   67: withVisible
   66: f
   65: Reduce
   64: withCallingHandlers
   63: domain$wrapSync
   62: promises::with_promise_domain
   61: captureStackTraces
   60: doTryCatch
   59: tryCatchOne
   58: tryCatchList
   57: tryCatch
   56: do
   55: hybrid_chain
   54: observeEvent(chat_tokens())
   53: contextFunc
   52: env$runWith
   51: withCallingHandlers
   50: domain$wrapSync
   49: promises::with_promise_domain
   48: captureStackTraces
   47: force
   46: domain$wrapSync
   45: promises::with_promise_domain
   44: withReactiveDomain
   43: domain$wrapSync
   42: promises::with_promise_domain
   41: ctx$run
   40: run
   39: withCallingHandlers
   38: domain$wrapSync
   37: promises::with_promise_domain
   36: captureStackTraces
   35: withCallingHandlers
   34: shinyCallingHandlers
   33: force
   32: withVisible
   31: withCallingHandlers
   30: domain$wrapSync
   29: promises::with_promise_domain
   28: captureStackTraces
   27: doTryCatch
   26: tryCatchOne
   25: tryCatchList
   24: tryCatch
   23: do
   22: hybrid_chain
   21: flushCallback
   20: FUN
   19: lapply
   18: ctx$executeFlushCallbacks
   17: .getReactiveEnvironment()$flush
   16: flushReact
   15: serviceApp
   14: ..stacktracefloor..
   13: withCallingHandlers
   12: domain$wrapSync
   11: promises::with_promise_domain
   10: captureStackTraces
    9: ..stacktraceoff..
    8: shiny::runApp
    7: shiny::runGadget
    6: doTryCatch
    5: tryCatchOne
    4: tryCatchList
    3: tryCatch
    2: btw_app_from_client
    1: btw::btw_app
Warning: Unknown or uninitialised column: `role`.
Warning: Error in [: Can't subset rows with `tokens$role == "assistant"`.
✖ Logical subscript `tokens$role == "assistant"` must be size 1 or 4, not 0.
  105: <Anonymous>
  104: signalCondition
  103: signal_abort
  102: cnd_signal
  101: <Anonymous>
  100: signalCondition
   99: signal_abort
   98: cnd_signal
   97: stop_indicator_size
   96: <Anonymous>
   95: vec_as_location
   94: withCallingHandlers
   93: subclass_row_index_errors
   92: vectbl_as_row_location
   91: vectbl_as_row_index
   90: [.tbl_df
   89: [
   88: observe
   87: <observer:observeEvent(chat_tokens())>
   86: ..stacktraceon..
   85: valueFunc
   84: ..stacktraceon..
   83: contextFunc
   82: env$runWith
   81: withCallingHandlers
   80: domain$wrapSync
   79: promises::with_promise_domain
   78: captureStackTraces
   77: force
   76: domain$wrapSync
   75: promises::with_promise_domain
   74: withReactiveDomain
   73: domain$wrapSync
   72: promises::with_promise_domain
   71: ctx$run
   70: ..stacktraceoff..
   69: isolate
   68: func
   67: withVisible
   66: f
   65: Reduce
   64: withCallingHandlers
   63: domain$wrapSync
   62: promises::with_promise_domain
   61: captureStackTraces
   60: doTryCatch
   59: tryCatchOne
   58: tryCatchList
   57: tryCatch
   56: do
   55: hybrid_chain
   54: observeEvent(chat_tokens())
   53: contextFunc
   52: env$runWith
   51: withCallingHandlers
   50: domain$wrapSync
   49: promises::with_promise_domain
   48: captureStackTraces
   47: force
   46: domain$wrapSync
   45: promises::with_promise_domain
   44: withReactiveDomain
   43: domain$wrapSync
   42: promises::with_promise_domain
   41: ctx$run
   40: run
   39: withCallingHandlers
   38: domain$wrapSync
   37: promises::with_promise_domain
   36: captureStackTraces
   35: withCallingHandlers
   34: shinyCallingHandlers
   33: force
   32: withVisible
   31: withCallingHandlers
   30: domain$wrapSync
   29: promises::with_promise_domain
   28: captureStackTraces
   27: doTryCatch
   26: tryCatchOne
   25: tryCatchList
   24: tryCatch
   23: do
   22: hybrid_chain
   21: flushCallback
   20: FUN
   19: lapply
   18: ctx$executeFlushCallbacks
   17: .getReactiveEnvironment()$flush
   16: flushReact
   15: serviceApp
   14: ..stacktracefloor..
   13: withCallingHandlers
   12: domain$wrapSync
   11: promises::with_promise_domain
   10: captureStackTraces
    9: ..stacktraceoff..
    8: shiny::runApp
    7: shiny::runGadget
    6: doTryCatch
    5: tryCatchOne
    4: tryCatchList
    3: tryCatch
    2: btw_app_from_client
    1: btw::btw_app

I was hoping a more specific question (and removing any R markdown formatting) would remove the issue, but this seems like a different error?

[gadenbuie]

What version of ellmer are you using?

[mjfrigaard]

Hi @gadenbuie sorry for the delayed response!

packageVersion("ellmer")
[1] ‘0.4.0’

If it's helpful, I tried again today and received the following error:

Warning: ERROR: An error occurred in `chat_append_stream(id="chat-chat")`
Caused by error:
! parse error: premature EOF
                                       {"_intent": "Create expanded re
                     (right here) ------^
Warning in append_stream_task$invoke(client, "chat", input$chat_user_input) :
  ERROR: An error occurred when invoking the ExtendedTask.
Caused by error:
! parse error: premature EOF
                                       {"_intent": "Create expanded re
                     (right here) ------^

[gadenbuie]

Are you still getting the Can't subset rows with tokens$role == "user". error? You shouldn't be if you're using ellmer v0.4.0.

Unfortunately, I don't think there's much I can do unless you can share a complete, small reproducible example. There's just too many confounding variables (and unknowns) for me to be able to recreate the issue.

By the way, have you seen the btw.md feature? It makes it a lot easier to store the system prompt and choose the client's provider and model without having to edit an .Rprofile or set a global R option.

[mjfrigaard]

The error is different now (I've been using the same prompt from above), but I will create a smaller Shiny app/R package with a downloadable R Markdown report so it can be reproduced.

Yes--I love the btw.md feature! In fact, that's the section that directly precedes this issue

[mjfrigaard]

@gadenbuie I've create a tiny package here: https://github.com/mjfrigaard/btwapp

I've documented the error in the README.md.

It seems to get through phase 1 just fine, but has trouble with phase 2:

[gadenbuie]

Thanks @mjfrigaard. Is that a private repo? I get a 404 when I try to open it.

[mjfrigaard]

Sorry @gadenbuie --it was. It's public now.

[gadenbuie]

Thanks for putting together the whole repo! It turns out this is the problem I fixed in #121. We had a small bit of unfortunate timing in that ellmer v0.4.0 came out just after I released btw but it changed how token counts are shared.

You can either downgrade back to ellmer v0.3.0 or install btw from GitHub. I'm hoping to get a btw release out relatively soon, once I've finished up the tool to run R code.