feat(chat-with-content): add tested helpers and dedicated CI

.github/workflows/chat-with-content.yml

@@ -0,0 +1,81 @@
+name: Chat with Content
+
+on:
+  workflow_call:
+
+# Setup the environment with the extension name for easy re-use
+# Also need the GH_TOKEN for the release-extension action to be able to use gh
+env:
+  EXTENSION_NAME: chat-with-content
+  GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+
+jobs:
+  extension:
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: ./extensions/${{ env.EXTENSION_NAME }}
+
+    steps:
+      # Checkout the repository so the rest of the actions can run with no issue
+      - uses: actions/checkout@v4
+
+      # We want to fail quickly if the linting fails, do that first
+      - uses: ./.github/actions/lint-extension
+        with:
+          extension-name: ${{ env.EXTENSION_NAME }}
+
+      # Run the Python tests before packaging. pyproject-file only picks the uv
+      # version, so pin the interpreter separately: without it uv resolves the
+      # newest Python satisfying requires-python, and the tests would stop matching
+      # the version Connect deploys on (manifest.json's python.version).
+      # `uses` steps ignore the job's working-directory, so this can't read the
+      # extension's .python-version; it preinstalls the same version that file
+      # pins for `uv run` below. Keep the two in step.
+      - uses: astral-sh/setup-uv@v9.0.0
+        with:
+          pyproject-file: ./extensions/${{ env.EXTENSION_NAME }}/pyproject.toml
+          python-version: "3.12"
+
+      - name: Run tests
+        run: uv run pytest
+
+      # Upload only the files the extension needs to run, leaving out tests,
+      # pyproject.toml, and other repo-only files.
+      - name: Upload extension files
+        uses: actions/upload-artifact@v4
+        with:
+          name: ${{ env.EXTENSION_NAME }}
+          path: |
+            extensions/${{ env.EXTENSION_NAME }}/app.py
+            extensions/${{ env.EXTENSION_NAME }}/helpers.py
+            extensions/${{ env.EXTENSION_NAME }}/requirements.txt
+            extensions/${{ env.EXTENSION_NAME }}/manifest.json
+
+      # Package up the extension into a TAR using the generalized
+      # package-extension action
+      - uses: ./.github/actions/package-extension
+        with:
+          extension-name: ${{ env.EXTENSION_NAME }}
+          artifact-name: ${{ env.EXTENSION_NAME }}
+
+  connect-integration-tests:
+    needs: extension
+    uses: ./.github/workflows/connect-integration-tests.yml
+    secrets: inherit
+    with:
+      extensions: '["chat-with-content"]'  # JSON array format to match the workflow input schema
+
+  release:
+    runs-on: ubuntu-latest
+    needs: [extension, connect-integration-tests]
+    # Release the extension using the release-extension action
+    # Will only create a GitHub release if merged to `main` and the semver
+    # version has been updated
+    steps:
+      # Checkout the repository so the rest of the actions can run with no issue
+      - uses: actions/checkout@v4
+
+      - uses: ./.github/actions/release-extension
+        with:
+          extension-name: ${{ env.EXTENSION_NAME }}

.github/workflows/extensions.yml

@@ -56,7 +56,6 @@ jobs:
             stock-report: extensions/stock-report/**
             simple-mcp-server: extensions/simple-mcp-server/**
             simple-shiny-chat-with-mcp: extensions/simple-shiny-chat-with-mcp/**
-            chat-with-content: extensions/chat-with-content/**
             pqr: extensions/pqr/**
 
       # When infra changed, test all simple extensions; otherwise use the filter output
@@ -68,6 +67,12 @@ jobs:
             echo "Infrastructure changed — testing all simple extensions"
             changes=$(find extensions -maxdepth 1 -mindepth 1 -type d -printf '%f\n' \
               | while read -r ext; do
+                  # Skip complex extensions: they have their own dedicated
+                  # workflow (which builds them) and would fail the simple
+                  # no-build flow (e.g. a missing frontend dist/).
+                  if [ -f ".github/workflows/$ext.yml" ]; then
+                    continue
+                  fi
                   if jq -e '.extension.minimumConnectVersion' "extensions/$ext/manifest.json" > /dev/null 2>&1; then
                     echo "$ext"
                   fi
@@ -212,6 +217,7 @@ jobs:
       package-vulnerability-scanner: ${{ steps.resolve.outputs.package-vulnerability-scanner }}
       runtime-version-scanner: ${{ steps.resolve.outputs.runtime-version-scanner }}
       usage-metrics-dashboard: ${{ steps.resolve.outputs.usage-metrics-dashboard }}
+      chat-with-content: ${{ steps.resolve.outputs.chat-with-content }}
 
     steps:
       - uses: actions/checkout@v4
@@ -233,6 +239,7 @@ jobs:
             package-vulnerability-scanner: extensions/package-vulnerability-scanner/**
             runtime-version-scanner: extensions/runtime-version-scanner/**
             usage-metrics-dashboard: extensions/usage-metrics-dashboard/**
+            chat-with-content: extensions/chat-with-content/**
 
       # When infra changed, trigger all complex extensions too
       - id: resolve
@@ -243,11 +250,13 @@ jobs:
             echo "package-vulnerability-scanner=true" >> $GITHUB_OUTPUT
             echo "runtime-version-scanner=true" >> $GITHUB_OUTPUT
             echo "usage-metrics-dashboard=true" >> $GITHUB_OUTPUT
+            echo "chat-with-content=true" >> $GITHUB_OUTPUT
           else
             echo "publisher-command-center=${{ steps.changes.outputs.publisher-command-center }}" >> $GITHUB_OUTPUT
             echo "package-vulnerability-scanner=${{ steps.changes.outputs.package-vulnerability-scanner }}" >> $GITHUB_OUTPUT
             echo "runtime-version-scanner=${{ steps.changes.outputs.runtime-version-scanner }}" >> $GITHUB_OUTPUT
             echo "usage-metrics-dashboard=${{ steps.changes.outputs.usage-metrics-dashboard }}" >> $GITHUB_OUTPUT
+            echo "chat-with-content=${{ steps.changes.outputs.chat-with-content }}" >> $GITHUB_OUTPUT
           fi
 
   # Creates and releases the Publisher Command Center extension using a custom
@@ -287,6 +296,15 @@ jobs:
     uses: ./.github/workflows/usage-metrics-dashboard.yml
     secrets: inherit
 
+  # Creates and releases the Chat with Content extension using a custom workflow
+  chat-with-content:
+    needs: [complex-extension-changes]
+    # Only runs if the `complex-extension-changes` job detects changes in the
+    # chat-with-content extension directory
+    if: ${{ needs.complex-extension-changes.outputs.chat-with-content == 'true' }}
+    uses: ./.github/workflows/chat-with-content.yml
+    secrets: inherit
+
   # All extensions have been linted, packaged, and released, if necessary
   # Continuing to update the extension list with the latest release data
 
@@ -300,6 +318,7 @@ jobs:
       package-vulnerability-scanner,
       runtime-version-scanner,
       usage-metrics-dashboard,
+      chat-with-content,
     ]
     permissions:
       pull-requests: write
@@ -383,6 +402,7 @@ jobs:
       package-vulnerability-scanner,
       runtime-version-scanner,
       usage-metrics-dashboard,
+      chat-with-content,
     ]
     if: ${{ always() }}
     outputs:

extensions/chat-with-content/.gitignore

@@ -1,4 +1,3 @@
 rsconnect-python
 .venv
 .env
-.python-version

extensions/chat-with-content/.python-version

@@ -0,0 +1 @@
+3.12

extensions/chat-with-content/CHANGELOG.md

@@ -5,6 +5,15 @@ All notable changes to the Chat with Content extension will be documented in thi
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.0.8] - 2026-07-17
+
+### Fixed
+
+- Guarded against content with a missing, malformed, or timezone-naive deployment
+  time, which could previously leave the whole content list empty. (#446)
+- Close a code block that truncation cut open, so the model reads the truncation
+  note as a note rather than as more code. (#446)
+
 ## [0.0.7] - 2026-06-15
 
 ### Changed

extensions/chat-with-content/helpers.py

@@ -1,31 +1,108 @@
+import os
 from datetime import datetime, timezone
 
+# Static/rendered content the app can extract text from. Interactive apps (Shiny,
+# Streamlit, ...) render in the browser, so their HTML holds no content to chat with.
+CHATTABLE_APP_MODES = ("jupyter-static", "quarto-static", "rmd-static", "static")
 
-def time_since_deployment(deployment_time_str):
-    """
-    Calculate time since deployment from ISO format datetime string.
+# Upper bound on the markdown handed to the LLM. A large report can produce a DOM
+# far bigger than the model's context window; truncating keeps the request within
+# bounds instead of erroring, at the cost of dropping the tail of very long pages.
+MAX_CONTEXT_CHARS = 100_000
+
+
+# Both env vars are checked because a missed "on Connect" detection would fall back
+# to the deploy client for a viewer (see resolve_visitor_client), so err toward True.
+def running_on_connect():
+    return "CONNECT" in (os.getenv("POSIT_PRODUCT"), os.getenv("RSTUDIO_PRODUCT"))
+
+
+# What the viewer is told when their session can't be used at all. Adding the
+# integration fixes neither case, so these are kept separate from the setup screen.
+# Self-contained sentences: the underlying error, if any, goes to the server log
+# instead (see resolve_visitor_client), since a viewer can't act on it and it may
+# be full of vendor/SDK detail that isn't meant for them.
+NO_SESSION_DETAIL = (
+    "Couldn't read your Connect session, so the app can't list or read content as "
+    "you. Make sure you're signed in to Connect. If you are, your administrator may "
+    "need to enable OAuth integrations on this server."
+)
+EXCHANGE_FAILED_DETAIL = (
+    "Couldn't read your Connect session, so the app can't list or read content as "
+    "you. Contact your administrator; the technical detail is in the application logs."
+)
+# Raised by app.py, not resolve_visitor_client: the exchange itself has no timeout,
+# so app.py bounds the call with asyncio.wait_for and reports this on expiry.
+SESSION_TIMEOUT_DETAIL = (
+    "Couldn't read your Connect session, so the app can't list or read content as "
+    "you. Connect didn't respond in time; try reloading the page."
+)
 
-    Args:
-        deployment_time_str (str): Datetime string in format "2025-03-19T23:16:11Z"
 
-    Returns:
-        str: Human-readable time difference like "last deployed 3 hours ago"
-    """
-    # Parse the deployment time
-    deployment_time = datetime.fromisoformat(deployment_time_str.replace("Z", "+00:00"))
+# Returns (client, integration_enabled, session_error), where session_error is the
+# detail to show the viewer when their session can't be used, or None otherwise. The
+# gate is the point: never fall back to the deploy client for a viewer, because that
+# would list the deployer's content as if it were theirs. Off Connect, the deploy
+# client is the intended one.
+def resolve_visitor_client(client, on_connect, token):
+    if not on_connect:
+        return client, True, None
+    if not token:
+        # No token means there is no signed-in viewer to act as: content that allows
+        # anonymous access sends none, and OAuth integrations may be off server-wide.
+        # Neither is fixed on the Access tab, so say that rather than showing setup.
+        return client, True, NO_SESSION_DETAIL
+    try:
+        return client.with_user_session_token(token), True, None
+    except Exception as err:
+        # Compare as a string so a code reported as 212 or "212" both count as the
+        # missing-integration case (setup screen) rather than a scary error screen.
+        if str(getattr(err, "error_code", "")) == "212":
+            return client, False, None
+        # The raw error may be full of SDK/vendor detail a viewer can't act on, so it
+        # goes to the log for an administrator rather than onto the screen.
+        raw = getattr(err, "error_message", None) or str(err)
+        print(f"chat-with-content: session token exchange failed: {raw}")
+        return client, True, EXCHANGE_FAILED_DETAIL
 
-    # Get current time in UTC
+
+# Whether the app is fully set up and should load and use the viewer's content.
+# The content selector runs as a reactive effect independent of the rendered
+# screen, so it gates on this itself: a token error leaves the client as the
+# unscoped deploy client (must never load with it), and a missing LLM or
+# integration means the setup screen is up, so there's nothing to load for yet.
+def content_ready(token_error, chat, integration_enabled):
+    return token_error is None and chat is not None and integration_enabled
+
+
+def time_since_deployment(deployment_time_str):
+    # Content that has never been deployed reports no time; skip the label rather
+    # than crash on a None passed to fromisoformat().
+    if not deployment_time_str:
+        return ""
+
+    try:
+        deployment_time = datetime.fromisoformat(
+            deployment_time_str.replace("Z", "+00:00")
+        )
+    except (AttributeError, TypeError, ValueError):
+        # A malformed timestamp shouldn't crash the whole content list; just omit
+        # the "last deployed" phrase for this one item. AttributeError covers a
+        # value that isn't a string at all, which has no .replace().
+        return ""
+    # A timestamp with no offset would raise when subtracted from an aware "now";
+    # treat it as UTC so a naive-but-valid time still renders instead of crashing.
+    if deployment_time.tzinfo is None:
+        deployment_time = deployment_time.replace(tzinfo=timezone.utc)
     current_time = datetime.now(timezone.utc)
 
-    # Calculate the difference
     time_diff = current_time - deployment_time
     total_seconds = time_diff.total_seconds()
 
-    # Handle future dates
+    # Deployment time slightly in the future (clock skew between servers).
     if total_seconds < 0:
         return "last deployed in the future"
 
-    # Convert to appropriate unit
     if total_seconds < 60:
         value = int(total_seconds)
         unit = "second" if value == 1 else "seconds"
@@ -49,3 +126,46 @@ def time_since_deployment(deployment_time_str):
         unit = "year" if value == 1 else "years"
 
     return f"last deployed {value} {unit} ago"
+
+
+# Content items are read with .get() throughout: Connect leaves optional fields out
+# of the payload, and the SDK raises AttributeError for a field accessed as an
+# attribute but absent, which would take out the whole content list over one unusual
+# item. Attribute access is also deprecated in the SDK in favour of key access.
+def is_chattable_content(item):
+    return (
+        item.get("app_mode") in CHATTABLE_APP_MODES
+        and item.get("app_role") != "none"
+        and item.get("content_category") != "pin"
+    )
+
+
+def content_choice_label(item):
+    title = item.get("title") or item.get("name") or item.get("guid")
+    # .get() also sidesteps ContentItem.owner, a property that fetches the owner
+    # over HTTP (one request per item) when Connect didn't include it.
+    owner = item.get("owner") or {}
+    name = f"{owner.get('first_name') or ''} {owner.get('last_name') or ''}".strip()
+    deployed = time_since_deployment(item.get("last_deployed_time"))
+
+    # Join only the parts we actually have so the label never shows " -  ".
+    suffix = " ".join(part for part in (name, deployed) if part)
+    return f"{title} - {suffix}" if suffix else title
+
+
+def truncate_for_context(markdown, max_chars=MAX_CONTEXT_CHARS):
+    if len(markdown) <= max_chars:
+        return markdown
+    kept = markdown[:max_chars]
+    # Cutting mid-page can leave a code fence open, which would make the model read
+    # the notice below as more code instead of as a note about the content. Only a
+    # fence starting a line opens or closes a block: page content that merely
+    # mentions ``` mid-line must not be counted, or a balanced block would be
+    # "closed" again and the notice pushed inside the new one.
+    fences = sum(1 for line in kept.splitlines() if line.lstrip().startswith("```"))
+    if fences % 2:
+        kept += "\n```"
+    return (
+        kept
+        + "\n\n[Content truncated because it exceeds the size this app sends to the model.]"
+    )

extensions/chat-with-content/manifest.json

@@ -24,29 +24,27 @@
     "description": "Provides a way to interact with and query content using an LLM chat interface.",
     "homepage": "https://github.com/posit-dev/connect-extensions/tree/main/extensions/chat-with-content",
     "category": "extension",
-    "tags": ["llm", "shiny", "chat", "python"],
+    "tags": [
+      "llm",
+      "shiny",
+      "chat",
+      "python"
+    ],
     "minimumConnectVersion": "2025.04.0",
-    "requiredFeatures": ["OAuth Integrations"],
+    "requiredFeatures": [
+      "OAuth Integrations"
+    ],
     "version": "0.0.7"
   },
   "files": {
     "requirements.txt": {
       "checksum": "0c56ba1ce838560d153c50e71dc2b025"
     },
-    ".gitignore": {
-      "checksum": "693ec79eaa892babde62587aaacf0d8b"
-    },
-    "README.md": {
-      "checksum": "7b072eef923c062a0bf1667dde6c2b95"
-    },
     "app.py": {
       "checksum": "74b2cad1b559b06c306a1ee44ce00185"
     },
     "helpers.py": {
-      "checksum": "b18f4bc0072b6e47864670a3174b0cea"
-    },
-    "pyproject.toml": {
-      "checksum": "604c28539dcb8abf952d7099ce2994bc"
+      "checksum": "cf9684fe4097503d78c69724e795579b"
     }
   }
 }