[devin-ai-integration]

Summary

Adds hermetic unit tests for the episode HTML export, AgentXray's non-UI helpers, and overlay utils. Only three new test files; no src/ or existing test changes.

• tests/analyze/test_episode_to_html.py builds a synthetic exp dir in tmp_path via StepInfo.save_step_info + ExpResult, then checks exp_result_to_html output: base64 data-URI vs file:// images (embed_images), the open/closed flags on sections, <script> escaping in goal/think/action/chat, plus tool-call and message-type formatting.
• tests/analyze/test_agent_xray_helpers.py covers the dataframe helpers, ClickMapper, markdown/chat formatting, screenshots/gallery, the update_* tabs, submit_action (OpenAI is mocked), the selection callbacks, get_directory_contents, and profiling. An autouse fixture swaps in a fresh agent_xray.info for each test so nothing leaks between tests (safe under xdist and random order). Gradio is never launched.
• tests/analyze/test_overlay_utils_unit.py covers AST parsing of parse_function_calls, bid and x/y pair merging, the PIL overlays and linedashed, annotate_action, and create_colored_html.

Coverage (target modules)

Module	Before (main)	After
episode_to_html	0%	91% (only the __main__ block, lines 424-442, is uncovered)
agent_xray	0%	74% (mostly the Gradio run_gradio UI construction, lines 178-550)
overlay_utils	57%	98%

Tests: 159 passed, 7 xfailed (140 test functions, some parametrized). About 7s serially and about 10s with -n 4.

Strict-xfailed source bugs (not fixed here)

• episode_to_html.exp_result_to_html: task_name is put into the HTML without escaping.
• episode_to_html._format_chat_messages_like_xray: the chat message role is not HTML-escaped.
• agent_xray.Info.update_exp_result: there is no return after the result_df is None guard.
• agent_xray.dict_msg_to_markdown: dict messages with plain-string content crash because the code iterates over the string's characters.
• agent_xray.update_agent_info_html: when html_page is missing, the default is a list, which _page_to_iframe can't encode.
• agent_xray.get_episode_info: errors when obs has no goal_object (str(BaseMessage("", None))), even though the KeyError is caught deliberately.
• overlay_utils.create_colored_html: raw action text is inserted into the HTML without escaping.

Run

uv run pytest tests/analyze/test_episode_to_html.py tests/analyze/test_agent_xray_helpers.py tests/analyze/test_overlay_utils_unit.py -p no:cacheprovider -q   # add -n 4 for xdist

Devin-Org: engineering

Link to Devin session: https://devin-gtm.devinenterprise.com/sessions/f0c73caf2b6e4a29a47075802f0b5f32
Open in Devin Desktop: https://devin-gtm.devinenterprise.com/desktop/session/f0c73caf2b6e4a29a47075802f0b5f32?variant=devin
Requested by: @danigrim

---

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• Disable automatic comment, CI, and merge conflict monitoring

[devin-ai-integration]

🔍 Devin Review: 1 flag

Not posted on this PR by your GitHub settings — view it in Devin Review. (Configure)