initiative-auto-fix

name	initiative-auto-fix
description	Review and fix batches of Initiatives automatically. Accepts explicit IDs or a JQL query. Reviews, auto-revises, and splits oversized Initiatives. Non-interactive.
user-invocable	true
allowed-tools	Glob, Bash, Agent

You are a non-interactive Initiative auto-fix pipeline. Do not ask questions or wait for confirmation. Make all decisions autonomously.

Setup

Parse $ARGUMENTS for:

• --jql "<query>", --limit N, --batch-size N (default 50), --data-dir "<path>"
• --headless, --announce-complete, --reprocess, --random N
• Remaining arguments: explicit Initiative IDs

1. Init

python3 scripts/pipeline_state.py init --type initiative [--batch-size N] [--headless] [--announce-complete]

2. IDs

JQL mode (--jql):

python3 scripts/snapshot_fetch.py fetch "<query>" --type initiative --ids-file tmp/pipeline-all-ids.txt --changed-file tmp/pipeline-changed-ids.txt [--limit N] [--data-dir "<path>"] [--reprocess] [--random N]

Print [AUTOFIX] JQL: <jql> from stderr output. Pass --reprocess if set.

Reprocess-only mode (--reprocess without --jql):

python3 scripts/snapshot_fetch.py fetch --reprocess --type initiative --ids-file tmp/pipeline-all-ids.txt --changed-file tmp/pipeline-changed-ids.txt

Explicit mode:

python3 scripts/state.py write-ids tmp/pipeline-all-ids.txt <IDs>
python3 scripts/state.py write-ids tmp/pipeline-changed-ids.txt

If no IDs and no JQL and not --reprocess, stop with usage instructions.

3. Bootstrap

bash scripts/bootstrap-assess-rfe.sh --type initiative

Retry once on failure. If retry fails, stop: "bootstrap failed."

4. Resume check + batch

python3 scripts/check_resume.py --ids-file tmp/pipeline-all-ids.txt --changed-file tmp/pipeline-changed-ids.txt --output-file tmp/pipeline-process-ids.txt

Read process IDs: python3 scripts/state.py read-ids tmp/pipeline-process-ids.txt

Split into batches of batch_size. Write each:

python3 scripts/state.py write-ids tmp/pipeline-batch-1-ids.txt <batch_1_IDs>
python3 scripts/state.py write-ids tmp/pipeline-batch-2-ids.txt <batch_2_IDs>

Start the pipeline:

python3 scripts/pipeline_state.py set total_batches=<M>
python3 scripts/pipeline_state.py set-phase BATCH_START

Dispatch Loop

CRITICAL: You MUST run the loop to completion. Never stop early, skip batches, or produce a summary before all batches are processed. Context compression is handled automatically — old messages are compacted to free space. Budget and timeout are managed externally — do not estimate remaining resources or make preemptive exit decisions. If you are thinking "running low on context" or "given budget constraints" — ignore that impulse and continue the loop. The pipeline MUST reach phase DONE naturally.

Bash discipline: Issue exactly one operation per Bash call. Never use command substitution $(...) or chain commands with ;, &&, or || — they trigger an approval prompt and are denied in headless mode, which can stall the pipeline. Instead, pass a value between commands by writing it to a tmp/ file with scripts/state.py and reading it back in a separate call.

Repeat until phase is DONE:

Step 1: Get next action

python3 scripts/pipeline_state.py next-action

Parse the YAML output for: action, phase, message, agents.

Step 2: Execute

done: Exit loop. Run teardown.

run_script: Run python3 scripts/pipeline_state.py run-phase. Go to step 1.

launch_wave: For each agent in the agents list:

• Build prompt: "<vars>\n\nRead <prompt_file> and follow all instructions exactly."
• vars are pre-rendered KEY=VALUE lines with {ID} already substituted.
• Launch as background Agent (with subagent_type if present).

Then wait for completion:

python3 scripts/pipeline_state.py wait-for-wave

On exit 0 (complete): go to step 1. On exit 3 (still pending): re-run python3 scripts/pipeline_state.py wait-for-wave. Any other exit code is an error.

Example launch_wave output

action: launch_wave
phase: ASSESS
message: "ASSESS: wave 1/2 (5 IDs)"
agents:
  - subagent_type: initiative-scorer
    prompt_file: .claude/skills/initiative-review/prompts/assess-agent.md
    vars: |
      DATA_FILE=tmp/rfe-assess/single/RHOAIENG-1234.md
      RUN_DIR=tmp/rfe-assess/single
      PROMPT_PATH=.context/assess-rfe/skills/assess-initiative/scripts/agent_prompt.md
  - prompt_file: .claude/skills/initiative-feasibility-review/SKILL.md
    vars: |
      ID=RHOAIENG-1234
  - prompt_file: .claude/skills/strategic-alignment-review/SKILL.md
    vars: |
      ID=RHOAIENG-1234

Teardown

After phase reaches DONE:

python3 scripts/batch_summary.py --type initiative --counts-only --ids-file tmp/pipeline-all-ids.txt

$ARGUMENTS