| name | create-issue |
| description | Convert a PyAutoMind prompt file into a GitHub issue and register it — the Mind issue and registry primitive that start_dev delegates to. |
Turn a PyAutoMind/ prompt into a tracked GitHub issue and register it.
A PyAutoMind primitive — it owns the mechanical issue + registry write:
assemble the issue body, create the issue, register the task in active.md, move
the prompt to active/, and push Mind state. The reasoning (classify repos,
explore code, generate the plan) belongs to PyAutoBrain:
- Called by Brain.
$start-dev (/start_dev in Claude) does the triage/planning via the
Feature Agent and then delegates the issue write to this skill, passing the
primary repo, title, plan and suggested branch. Brain does not re-implement
issue creation.
- Runnable standalone. You can run
$create-issue (/create_issue in Claude) by itself when you just
want an issue from a prompt without full triage or worktree routing — it does a
light pass to fill any inputs the caller didn't supply. For deep
classification + dev-environment setup, use $start-dev instead.
Organ boundary and the execution-environment model: PyAutoBrain skills/WORKFLOW.md.
Usage
$create-issue <prompt-file-path> # /create_issue in Claude
Path relative to PyAutoMind/. Prompts live under <work-type>/<target>/ (see
README "Prompt taxonomy"); pre-migration <target>/<name>.md paths still resolve.
Examples: bug/autofit/factor_graph_instance_iteration.md,
feature/autoarray/oversampling.md.
Inputs
These come from the caller (Brain/$start-dev) when delegated, or from a
light standalone pass here when run directly:
| Input | From caller | Standalone fallback |
|---|
| primary repo | Feature Agent classification | most-referenced @RepoName in the prompt; ask if ambiguous |
| title | caller | concise title (<70 chars), conventional prefix (feat:/fix:/refactor:/docs:/perf:) |
| plan (high + detailed) | caller | brief plan from a quick read of the prompt + referenced files |
| suggested branch | caller (plan_branches) | feature/<short-desc> kebab-case, <50 chars |
Resolve repository owners from PyAutoMind/repos.yaml; the generated summary
in WORKFLOW.md is the readable mapping. Do not reuse legacy owner defaults.
Steps
0. Sync new prompt ideas (Mind)
source PyAutoMind/scripts/prompt_sync.sh
prompt_sync_new_prompts
1. Read the prompt
Read PyAutoMind/<argument>. If missing, report and list prompts in that folder.
2. Resolve the inputs
If the caller supplied repo/title/plan/branch, use them verbatim. Otherwise
do the light standalone pass from the Inputs table (don't run a full Brain
triage — that's $start-dev).
3. Assemble + create the issue
Build the body in this structure, then create it (present for review first):
## Overview
<2-4 sentence summary of what this task is and why it matters>
## Plan
<high-level bullet plan — human readable, no code>
<details>
<summary>Detailed implementation plan</summary>
### Affected Repositories
- repo1 (primary)
### Branch Survey
| Repository | Current Branch | Dirty? |
|-----------|---------------|--------|
| ./RepoName | main | clean |
**Suggested branch:** `feature/<name>`
### Implementation Steps
1. <step with file paths>
### Key Files
- `path/to/file.py` — description
</details>
## Original Prompt
<details>
<summary>Click to expand starting prompt</summary>
<original prompt content copied verbatim>
</details>
gh issue create --repo <owner/repo> --title "<title>" --body "$(cat <<'ISSUE_EOF'
<body content>
ISSUE_EOF
)"
4. Register the task in active.md (Mind)
Add the task entry to PyAutoMind/active.md with the issue URL (schema in
README). If the caller is handling registration itself — e.g. $start-dev
routing a conflicted task to planned.md — skip this step and let it register.
5. Move the prompt to active/
The prompt advances from draft/ (not started) to active/ (issued, in
flight) — the second of the three lifecycle states (draft/ → active/ → complete/; issue #71). Use git mv to preserve history:
mkdir -p PyAutoMind/active
git -C PyAutoMind mv <draft-path> active/<filename>
The mkdir -p matters on a freshly-spawned Mind, where active/ does not
yet exist (it holds only instance state, so the template ships without it) —
git mv errors if the destination dir is missing. Timestamp-suffix the
filename if one already exists in active/.
6. Push Mind
source PyAutoMind/scripts/prompt_sync.sh
prompt_sync_push "prompt: file issue for <task-name> (#<issue>)"
If step 0 already pushed, this carries only the active.md + active/ changes.
Notes
- Always present the issue body for review before creating it.
- If
gh auth status fails, tell the user to run ! gh auth login.
- The detailed plan should be thorough enough that a fresh session could start
from the issue alone.
