myna-dev-execute-tasks

name	myna-dev-execute-tasks
description	Run the Myna dev task queue — reads pending tasks from tmp/tasks.md, executes them sequentially on a single fix branch, each with its own implement→review→fix loop, then pushes the branch. Use when: "run the queue", "execute tasks", "process the queue", "work through the task list". Re-reads the queue before each task, so tasks added mid-run are picked up automatically.
user-invocable	true
allowed-tools	["Read","Write","Edit","Bash","Agent"]
effort	max

Myna Execute Tasks

You are the queue orchestrator. Your job: run every pending task in tmp/tasks.md sequentially on a single branch, each via a dedicated task subagent, then push the branch. You do not implement anything yourself — you sequence, coordinate, and track.

---

Step 1 — Read the Queue

Read tmp/tasks.md. If it doesn't exist:

tmp/tasks.md not found. Add tasks first with /myna-dev-task-add.

Stop.

If no pending tasks in the table:

No pending tasks in tmp/tasks.md.

Stop.

List what you found:

Found [N] pending task(s):
- Task [#]: [Title]
- Task [#]: [Title]

---

Step 2 — Create the Branch

Get today's date:

date +%Y-%m-%d

Try fix/[date]:

git checkout -b fix/[date] 2>/dev/null && echo "ok" || echo "exists"

If exists, try fix/[date]-2, fix/[date]-3, etc. until one succeeds.

Record the branch name — all tasks run on this branch.

---

Step 3 — Set Up the Run Log

Create the run folder and log:

mkdir -p tmp/[date]/reviews

Write tmp/[date]/run.md:

# Queue run — [date]

**Branch:** fix/[date]

## Tasks
| # | Title | Status | Review rounds | Reports |
|---|---|---|---|---|
| [#] | [Title] | pending | — | — |
| [#] | [Title] | pending | — | — |

## Notes
(none yet)

---

Step 4 — Execute Tasks

Process pending tasks in order (lowest task number first — FIFO).

Before each task: re-read tmp/tasks.md. This picks up any tasks added while the queue is running. If new pending tasks appeared below the current position, they will be processed in subsequent iterations of this loop.

For each pending task:

4a. Derive the short-name

From the task title, extract a 2-3 word identifier: lowercase, hyphen-separated, drop articles and prepositions. Used for review file naming.

Examples:

• "Add base guard to calendar skill" → base-guard
• "Fix email triage sort order" → email-sort
• "Update vault path in meeting skill" → meeting-path
• "Remove duplicate steering logic from capture" → capture-steering

4b. Record the base commit

git rev-parse HEAD

Save this SHA as BEFORE_SHA for this task. This scopes the review diff to only this task's changes, not everything on the branch.

4c. Update run log — mark in progress

Update the task's row in tmp/[date]/run.md to in-progress.

4d. Spawn the task subagent

Use the Agent tool. Prompt:

You are a Myna dev task subagent. Implement the task below, then invoke /myna-dev-task-protocol to commit, review, fix, and push.

---
[paste the full task body from tmp/tasks.md — the entire ## Task [N] — [Title] section]
---

After implementing, invoke:
/myna-dev-task-protocol --branch fix/[date] --base [BEFORE_SHA] --criteria "[done-when assertions as comma-separated string]" --short-name [short-name] --feature [date] --commit-msg "[type]([scope]): [description]"

Branch fix/[date] already exists and you are already on it — do not create it.

Report back exactly as implement-task instructs: "Done — [N] round(s) — [paths]" or "Unresolved: [list] — [paths]".

Replace [done-when assertions] with the task's "Done when" bullets joined with ", ".

Replace [type]([scope]): [description] with a conventional commit message appropriate for the task. Use fix for bug fixes, feat for new behavior, docs for doc-only changes. Scope is the skill or area being changed.

Wait for the subagent to complete and report back.

4e. Update after completion

If "Done":

• Update tmp/tasks.md table row: status → done, fill in review rounds and report paths
• Update tmp/[date]/run.md table row: same
• Changelog: if the task's changelog field is yes and a changelog-line field is present, append that line to the [Unreleased] section in CHANGELOG.md at the repo root. Read the current [Unreleased] block, append the line under it, and write back. Do this once per task at completion — not per commit within the task. Skip silently if changelog: no or the field is absent.
• Continue to the next task

If "Unresolved":

• Update tmp/tasks.md table row: status → failed, fill in report paths
• Update tmp/[date]/run.md table row: same; add to Notes section: "Task [#] failed: [unresolved issues]"
• Continue to the next task — do not block the queue

If subagent errors or does not complete:

• Update status to error in both files
• Note in run log
• Continue

---

Step 5 — Write the Summary

After all pending tasks have been processed, write tmp/[date]/summary.md:

# Queue run — [date]

**Branch:** fix/[date]
**Tasks processed:** [N]

## Results

| # | Title | Result | Review rounds | Reports |
|---|---|---|---|---|
| [#] | [Title] | done | [N] | [short-name]-r1.md[, -r2.md] |
| [#] | [Title] | failed | [N] | [short-name]-r1.md, ... |

## Review findings
[Per task: what was found in r1, what was resolved, what remained]

## Notes
[Any deviations, files changed beyond suggestions, judgment calls]

---

Step 5b — Archive Completed Tasks

After writing the summary, move all done tasks out of tmp/tasks.md. Failed and pending tasks stay — they represent work still to do.

Determine archive filename:

mkdir -p tmp/archive

Start with tmp/archive/tasks-[date].md. If it exists, try tasks-[date]-2.md, tasks-[date]-3.md, etc. until one is free.

Write the archive file using the same format as tasks.md — table header, then one row per done task, then the full ## Task N — [Title] body sections for each done task. Include a header line at the top:

# Archived tasks — [date] run

[table with done task rows only]

[full body sections for done tasks only]

Strip done tasks from tmp/tasks.md:

• Remove each done task's row from the table
• Remove each done task's full ## Task N — [Title] body section
• Leave the table header, pending rows, failed rows, and their body sections intact
• If no tasks remain after stripping, leave just the table header with no rows

Do not renumber remaining tasks — preserve original task numbers.

---

Step 6 — Push and Report

Push the branch:

git push origin fix/[date]

Report results:

Queue run complete — [N] done[, [M] failed].

Branch: fix/[date]

Tasks:
- Task [#]: [Title] — done ([N] review round(s))
- Task [#]: [Title] — failed — [unresolved issues]

[If any failures:]
Failed tasks remain in tmp/tasks.md — fix them and run the queue again, or remove them if no longer needed.

Summary: tmp/[date]/summary.md

---

What You Do Not Do

• Do not implement tasks yourself — every task goes to a subagent.
• Do not create per-task branches — all tasks run on fix/[date].
• Do not create PRs — push the branch and report; merging is the user's decision.
• Do not stop the queue when a task fails — continue with the remaining tasks.
• Do not re-read the queue at the start and never again — re-read before each task.