remote

Remote

Polls ~/nickolashkraus/agent-os/master/tasks/remote/ for task files submitted from Obsidian on iOS. Each task is a Markdown file with a prompt and metadata. Run repeatedly for continuous polling.

Usage

$remote              Poll once and execute any pending tasks.
$remote poll         Same as above.
$remote list         List all tasks and their statuses.
$remote clean        Remove completed and failed task files.

Continuous polling (self-paced):

Run repeatedly: $remote

Task format

Each task is a Markdown file in tasks/remote/. The filename is the task slug (e.g., deploy-staging.md, fix-typo-in-readme.md).

---
status: pending
created: 2026-04-27T10:30:00
project: nickolashkraus
---

The prompt for Codex goes here. Write it exactly as you would
type it into Codex.

Fields

• status: pending, in-progress, blocked, done, failed.
• created: ISO 8601 timestamp.
• project: The directory name under ~/nickolashkraus/ (or ~/Function-Health/) where the task should be executed.

Blocked tasks

When an agent needs clarification, it sets status: blocked and appends a ## Blocked section with the question:

---
status: blocked
created: 2026-04-27T10:30:00
project: nickolashkraus
---

The original prompt.

## Blocked

Which database should I run the migration against, Dev or Prod?

To unblock, add an ## Answer section and change the status back to pending:

---
status: pending
created: 2026-04-27T10:30:00
project: nickolashkraus
---

The original prompt.

## Blocked

Which database should I run the migration against, Dev or Prod?

## Answer

Dev.

The next poll picks it up and continues with the original prompt plus the Q&A context.

Result

When a task completes, the skill updates the frontmatter and appends a result block:

---
status: done
created: 2026-04-27T10:30:00
completed: 2026-04-27T10:35:00
project: nickolashkraus
---

The original prompt.

## Result

A concise summary of what was done.

Failed tasks get status: failed and a ## Result block explaining the error.

Step 1: Parse arguments

Parse the user-provided skill input to determine the subcommand. Default to poll if no arguments are provided.

Step 2: Execute subcommand

poll (default)

1. List all .md files in tasks/remote/.
2. Filter for files with status: pending in the frontmatter.
3. Sort by created timestamp (oldest first).
4. For each pending task: a. Update status to in-progress. b. Read the project field to determine the working directory.
   • If project starts with FH: or fh:, use ~/Function-Health/<rest>.
   • Otherwise, use ~/nickolashkraus/<project>/master (assumes worktree layout). If that path does not exist, try ~/nickolashkraus/<project>. c. Extract the full file content after the frontmatter closing ---. This includes the original prompt and, for unblocked tasks, any ## Blocked and ## Answer sections that provide additional context. d. Execute the task by spawning a Codex subagent in the target project directory. Pass the full content to the agent. e. If the agent needs clarification and cannot proceed:
   • Set status to blocked.
   • Append a ## Blocked section with the question.
   • Do not append a ## Result section. f. When the agent completes, update the task file:
   • Set status to done (or failed if the agent errored).
   • Add completed: <ISO 8601> to the frontmatter.
   • Append a ## Result section with the agent's summary.
5. After all tasks are processed, print a summary.

If no pending tasks exist, print:

No pending remote tasks.

list

Read all .md files in tasks/remote/ and print a table:

Remote tasks:

  # | Status      | Created              | File
  1 | done        | 2026-04-27T10:30:00  | deploy-staging.md
  2 | blocked     | 2026-04-27T10:45:00  | run-migration.md
  3 | in-progress | 2026-04-27T11:00:00  | fix-typo.md
  4 | pending     | 2026-04-27T11:15:00  | update-deps.md

clean

Delete all task files with status: done or status: failed. Print the number of files removed.

Step 3: Schedule next poll

This skill is designed for a self-paced recurring Codex run rather than a fixed interval. After each poll, schedule the next run with adaptive timing:

• Active: If any task was executed, blocked, or unblocked this poll, schedule the next poll in 60 seconds. There may be follow-up work or an answer arriving soon.
• Idle: If no tasks were processed (nothing pending, nothing newly unblocked), schedule the next poll in 270 seconds. This keeps the cache warm without wasting cycles.

~/.codex/rules/meta-learning.md