feat: add private notes note-taking agent by Ju-usc · Pull Request #231 · openhome-dev/abilities

Ju-usc · 2026-03-29T23:10:03Z

Summary

add a new private-notes voice-first note-taking agent under community/private-notes
implement a single-prompt tool loop for create/read/update/delete note flows with confirmation and safe JSON persistence
document the behavior, storage model, and validation steps in the ability README and SPEC

Validation

python3 -m py_compile abilities/community/private-notes/main.py
python3 abilities/validate_ability.py abilities/community/private-notes

Introduce a WebUI-compatible ability that relays coding tasks to a Codex webhook with confirmation/cancel flow and conversational summaries. Document setup, trigger words, and a minimal Codex-focused /run webhook contract for contributors.

Replace explicit register_capability boilerplate with the template register tag and remove raw open() usage so the ability matches updated validation requirements.

- Rename folder, class, log tag, and voice prompts to be agent-agnostic - Add _refine_prompt to clean up STT transcription before sending to agent - README shows Claude Code and Codex equally with unified webhook example - End-to-end tested via OpenHome + ngrok + Claude Code webhook

github-actions · 2026-03-29T23:10:12Z

✅ Community PR Path Check — Passed

All changed files are inside the community/ folder. Looks good!

github-actions · 2026-03-29T23:10:13Z

🔀 Branch Merge Check

PR direction: feat/private-notes-agent → dev

✅ Passed — feat/private-notes-agent → dev is a valid merge direction

github-actions · 2026-03-29T23:10:16Z

✅ Ability Validation Passed

📋 Validating: community/coding-agent-runner
  ✅ All checks passed!

📋 Validating: community/private-notes
  ✅ All checks passed!

github-actions · 2026-03-29T23:10:20Z

🔍 Lint Results

🔧 Auto-formatted

Some files were automatically cleaned and formatted with autoflake + autopep8 and committed.

✅ Unused imports removed (autoflake)
✅ Unused variables removed (autoflake)
✅ PEP8 formatting applied (autopep8)

✅ `init.py` — Empty as expected

Files linted: community/coding-agent-runner/main.py community/private-notes/main.py

✅ Flake8 — Passed

✅ All checks passed!

Copilot

Pull request overview

Adds a new voice-first, file-persisted private note-taking ability using an LLM-driven tool loop, plus introduces a separate “coding agent runner” webhook ability.

Changes:

Added private-notes ability with CRUD note tools, confirmation gates, and JSON persistence.
Added documentation for private-notes (README + SPEC).
Added a new coding-agent-runner ability (webhook-driven coding task execution) and docs.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 7 comments.

Show a summary per file

File	Description
community/private-notes/main.py	Implements tool-loop note CRUD + persistence for private notes.
community/private-notes/README.md	Documents usage, UX, and storage model for private notes.
community/private-notes/SPEC.md	Defines architecture, data model, tools, and safety rules for private notes.
community/private-notes/init.py	Package marker for the new ability.
community/coding-agent-runner/main.py	Adds a webhook-based coding task execution ability.
community/coding-agent-runner/README.md	Documents setup and webhook contract for coding agent runner.
community/coding-agent-runner/init.py	Package marker for the new ability.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-03-29T23:13:32Z

community/private-notes/main.py

+        notes = sorted(
+            notebook["notes"], key=lambda n: n.get("updated_at", ""), reverse=True
+        )
+        note_index = {
+            "note_count": len(notes),
+            "notes": [
+                {"id": n.get("id"), "title": n.get("title"), "updated_at": n.get("updated_at")}
+                for n in notes
+            ],
+        }


The note index included in the initial context is unbounded (it includes every note's id/title/updated_at). As the user accumulates notes, this can exceed the model context window and make the ability fail or behave unpredictably. Consider capping the index to a fixed number of most-recent notes (and/or adding a search/list tool to retrieve ids on demand).

Copilot · 2026-03-29T23:13:33Z

community/private-notes/main.py

+            notebook = await self._load_notebook()
+
+            # Capture time once so the context prefix stays identical across turns (LLM caching).
+            now = datetime.now(ZoneInfo(self.capability_worker.get_timezone()))


ZoneInfo(self.capability_worker.get_timezone()) can raise if the timezone string is missing/invalid. Right now that would abort the whole ability via the outer except. Consider wrapping this in a small try/except and falling back to UTC or naive datetime.now() so the ability can still function.

Suggested change

now = datetime.now(ZoneInfo(self.capability_worker.get_timezone()))

try:

tz = ZoneInfo(self.capability_worker.get_timezone())

except Exception:

tz = None

now = datetime.now(tz) if tz is not None else datetime.now()

Copilot · 2026-03-29T23:13:33Z

community/private-notes/main.py

+        if not existing:
+            return {"ok": False, "notes_changed": False, "error": "note not found"}
+
+        if not await self.capability_worker.run_confirmation_loop(args.get("confirmation", "")):


Overwrite confirmation is driven by args.get("confirmation", ""), which can become an empty string if the model omits/returns null for confirmation. That risks a confusing/blank confirmation prompt (or unexpected behavior inside run_confirmation_loop). Consider validating that a non-empty confirmation prompt was provided for overwrites, and returning a structured error (or forcing an ask_followup) if it's missing.

Suggested change

if not await self.capability_worker.run_confirmation_loop(args.get("confirmation", "")):

confirmation_prompt = args.get("confirmation")

if not isinstance(confirmation_prompt, str) or not confirmation_prompt.strip():

return {

"ok": False,

"notes_changed": False,

"error": "missing or empty confirmation prompt for overwrite",

}

if not await self.capability_worker.run_confirmation_loop(confirmation_prompt):

Copilot · 2026-03-29T23:13:33Z

community/private-notes/main.py

+
+        if not await self.capability_worker.run_confirmation_loop(args.get("confirmation", "")):


Delete confirmation is also taken from args.get("confirmation", ""), which can be empty/null if the model misbehaves. Since delete is destructive, consider requiring a non-empty confirmation prompt and returning an error/asking follow-up if it’s missing rather than calling run_confirmation_loop with an empty string.

Suggested change

if not await self.capability_worker.run_confirmation_loop(args.get("confirmation", "")):

confirmation = (args.get("confirmation") or "").strip()

# Require a non-empty confirmation prompt before running the confirmation loop.

if not confirmation:

return {

"ok": False,

"notes_changed": False,

"deleted_count": 0,

"status": "missing_confirmation",

}

if not await self.capability_worker.run_confirmation_loop(confirmation):

Copilot · 2026-03-29T23:13:34Z

community/private-notes/SPEC.md

+## Validation
+
+```
+python3 -m py_compile abilities/community/private-notes/main.py
+python3 abilities/validate_ability.py abilities/community/private-notes
+```


The validation commands reference abilities/community/private-notes/..., but in this repo the path is community/private-notes/... (and validate_ability.py is at the repo root). This makes the instructions in the spec non-runnable as written; please update the paths to match the repository layout.

Copilot · 2026-03-29T23:13:34Z

community/coding-agent-runner/main.py

+"""OpenHome ability – voice-triggered coding task execution via webhook.
+
+Flow: ask → confirm → refine prompt → call webhook → speak result.
+"""
+
+import asyncio
+
+import requests
+from src.agent.capability import MatchingCapability
+from src.main import AgentWorker
+from src.agent.capability_worker import CapabilityWorker
+


PR metadata describes only adding the private-notes ability, but this change set also introduces a full new coding-agent-runner ability. If this is intentional, it should be called out explicitly in the PR title/description; otherwise, consider splitting into a separate PR to keep review and release scope clear.

Copilot · 2026-03-29T23:13:34Z

community/private-notes/main.py

+                    system_prompt=SYSTEM_PROMPT,
+                )
+                # LLMs sometimes wrap JSON in markdown fences
+                tool_call = json.loads(llm_response.replace("```json", "").replace("```", "").strip())


llm_response is parsed with json.loads(...) without handling JSONDecodeError. If the model returns non-JSON (or extra text), this will jump to the broad exception handler and abort the entire ability. Consider catching json.JSONDecodeError here, logging the raw response (careful about PII), and either retrying the LLM call or prompting the user/LLM with ask_followup to restate the request/tool call format.

Suggested change

tool_call = json.loads(llm_response.replace("```json", "").replace("```", "").strip())

sanitized_response = llm_response.replace("```json", "").replace("```", "").strip()

try:

tool_call = json.loads(sanitized_response)

except json.JSONDecodeError:

# Log the raw LLM response for debugging, being careful about PII.

self.worker.editor_logging_handler.warning(

"[PrivateNotes] Failed to parse LLM tool call JSON. Raw response (sanitized): %r",

sanitized_response,

)

await self.capability_worker.speak(

"I had trouble understanding that notes request. Could you say it again?"

)

followup = await self._get_user_input(

"I didn't catch anything, so I didn't change your notes."

)

if not followup:

return

history.append({"role": "user", "content": followup})

continue

Ju-usc · 2026-03-29T23:13:58Z

Closing this PR because it was opened from the wrong branch and included unrelated coding-agent-runner history. It is superseded by #232, which contains only the private-notes changes based directly on dev.

Ju-usc and others added 13 commits February 22, 2026 21:36

Add Codex Task Runner community ability

44e2db8

Introduce a WebUI-compatible ability that relays coding tasks to a Codex webhook with confirmation/cancel flow and conversational summaries. Document setup, trigger words, and a minimal Codex-focused /run webhook contract for contributors.

docs: align webhook example with safe scoped Codex execution

c4df336

fix: align Codex Task Runner with new validator rules

b367543

Replace explicit register_capability boilerplate with the template register tag and remove raw open() usage so the ability matches updated validation requirements.

style: auto-format Python files with autoflake + autopep8

9f4007a

fix: preserve register capability tag format

067a59f

style: auto-format Python files with autoflake + autopep8

fd1d78e

fix: address PR feedback on async webhook and docs hardening

a6125eb

fix: preserve register capability tag format

9a7dff3

style: auto-format Python files with autoflake + autopep8

32590ab

fix: keep validator register tag stable under auto-format

5146102

refactor: simplify Codex Task Runner flow

2796cfd

feat: add private notes note-taking agent

3d4d378

Ju-usc requested a review from a team as a code owner March 29, 2026 23:10

Copilot AI review requested due to automatic review settings March 29, 2026 23:10

github-actions bot added the community-ability Community-contributed ability label Mar 29, 2026

style: auto-format Python files with autoflake + autopep8

b29d557

Copilot started reviewing on behalf of Ju-usc March 29, 2026 23:10 View session

Copilot AI reviewed Mar 29, 2026

View reviewed changes

Ju-usc closed this Mar 29, 2026

-            now = datetime.now(ZoneInfo(self.capability_worker.get_timezone()))
+            try:
+                tz = ZoneInfo(self.capability_worker.get_timezone())
+            except Exception:
+                tz = None
+            now = datetime.now(tz) if tz is not None else datetime.now()


		if not await self.capability_worker.run_confirmation_loop(args.get("confirmation", "")):

-        if not await self.capability_worker.run_confirmation_loop(args.get("confirmation", "")):
+        confirmation = (args.get("confirmation") or "").strip()
+        # Require a non-empty confirmation prompt before running the confirmation loop.
+        if not confirmation:
+            return {
+                "ok": False,
+                "notes_changed": False,
+                "deleted_count": 0,
+                "status": "missing_confirmation",
+            }
+        if not await self.capability_worker.run_confirmation_loop(confirmation):

-                tool_call = json.loads(llm_response.replace("```json", "").replace("```", "").strip())
+                sanitized_response = llm_response.replace("```json", "").replace("```", "").strip()
+                try:
+                    tool_call = json.loads(sanitized_response)
+                except json.JSONDecodeError:
+                    # Log the raw LLM response for debugging, being careful about PII.
+                    self.worker.editor_logging_handler.warning(
+                        "[PrivateNotes] Failed to parse LLM tool call JSON. Raw response (sanitized): %r",
+                        sanitized_response,
+                    )
+                    await self.capability_worker.speak(
+                        "I had trouble understanding that notes request. Could you say it again?"
+                    )
+                    followup = await self._get_user_input(
+                        "I didn't catch anything, so I didn't change your notes."
+                    )
+                    if not followup:
+                        return
+                    history.append({"role": "user", "content": followup})
+                    continue

Conversation

Ju-usc commented Mar 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Validation

Uh oh!

github-actions bot commented Mar 29, 2026

✅ Community PR Path Check — Passed

Uh oh!

github-actions bot commented Mar 29, 2026

🔀 Branch Merge Check

Uh oh!

github-actions bot commented Mar 29, 2026

✅ Ability Validation Passed

Uh oh!

github-actions bot commented Mar 29, 2026

🔍 Lint Results

🔧 Auto-formatted

✅ __init__.py — Empty as expected

✅ Flake8 — Passed

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Mar 29, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 29, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 29, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 29, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 29, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 29, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 29, 2026

Choose a reason for hiding this comment

Uh oh!

Ju-usc commented Mar 29, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Ju-usc commented Mar 29, 2026 •

edited

Loading

✅ `init.py` — Empty as expected