Ralph

EXECUTE NOW

Target: $ARGUMENTS

Parse arguments:

• N (required unless --dry-run): number of tasks to process
• --parallel: concurrent claim workers (max 5) + cross-connect validation
• --batch [id]: process only tasks from specific batch
• --type [type]: process only tasks at a specific phase (extract, create, reflect, reweave, verify, enrich)
• --dry-run: show what would execute without running
• --handoff: output structured RALPH HANDOFF block at end (for pipeline chaining)

Step 0: Read Vocabulary

Read ops/derivation-manifest.md (or fall back to ops/derivation.md) for domain vocabulary mapping. All output must use domain-native terms. If neither file exists, use universal terms.

START NOW. Process queue tasks.

MANDATORY CONSTRAINT: SUBAGENT SPAWNING IS NOT OPTIONAL

You MUST use the Task tool to spawn a subagent for EVERY task. No exceptions.

This is not a suggestion. This is not an optimization you can skip for "simple" tasks. The entire architecture depends on fresh context isolation per phase. Executing tasks inline in the lead session:

• Contaminates context (later tasks run on degraded attention)
• Skips the handoff protocol (learnings are not captured)
• Violates the ralph pattern (one phase per context window)

If you catch yourself about to execute a task directly instead of spawning a subagent, STOP. Call the Task tool. Every time. For every task. Including create tasks. Including "simple" tasks.

The lead session's ONLY job is: read queue, spawn subagent, evaluate return, update queue, repeat.

Phase Configuration

Each phase maps to specific Task tool parameters. Use these EXACTLY when spawning subagents.

All phases use the same subagent configuration:

• subagent_type: knowledge-worker (if available) or default
• mode: dontAsk

Subagents inherit the session model. Users running opus get opus quality on processing phases. Users running sonnet get sonnet everywhere. Fresh context per phase already ensures efficiency — every phase gets full capability in the smart zone.

Step 1: Read Queue State

Read the queue file. Check these locations in order:

1. ops/queue.yaml
2. ops/queue/queue.yaml
3. ops/queue/queue.json

Parse the queue. Identify ALL pending tasks.

Queue structure (v2 schema):

The queue uses current_phase and completed_phases per task entry:

phase_order:
  claim: [create, reflect, reweave, verify]
  enrichment: [enrich, reflect, reweave, verify]

tasks:
  - id: source-name
    type: extract
    status: pending
    source: ops/queue/archive/2026-01-30-source/source.md
    file: source-name.md
    created: "2026-01-30T10:00:00Z"

  - id: claim-010
    type: claim
    status: pending
    target: "claim title here"
    batch: source-name
    file: source-name-010.md
    current_phase: reflect
    completed_phases: [create]

If the queue file does not exist or is empty, report: "Queue is empty. Use /seed or /pipeline to add sources."

Step 2: Filter Tasks

Build a list of actionable tasks — tasks where status == "pending". Order by position in the tasks array (first = highest priority).

Apply filters:

• If --batch specified: keep only tasks where batch matches
• If --type specified: keep only tasks where current_phase matches (e.g., --type reflect finds tasks whose current_phase is "reflect")

The phase_order header defines the phase sequence:

• claim: create -> reflect -> reweave -> verify
• enrichment: enrich -> reflect -> reweave -> verify

Step 3: If --dry-run, Report and Stop

Show this and STOP (do not process):

--=={ ralph dry-run }==--

Queue: X total tasks (Y pending, Z done)

Phase distribution:
  Claims:       {create: N, reflect: N, reweave: N, verify: N}
  Enrichments:  {enrich: N, reflect: N, reweave: N, verify: N}

Next tasks to process:
1. {id} — phase: {current_phase} — {target}
2. {id} — phase: {current_phase} — {target}
...

Estimated: ~{N} subagent spawns

Step 4: Process Loop (SERIAL MODE)

If --parallel is set, skip to Step 6 instead.

Process up to N tasks (default 1). For each iteration:

4a. Select Next Task

Pick the first pending task from the filtered list. Read its metadata: id, type, file, target, batch, current_phase, completed_phases.

The current_phase determines which skill to invoke.

Report:

=== Processing task {i}/{N}: {id} — phase: {current_phase} ===
Target: {target}
File: {file}

4b. Build Subagent Prompt

Construct a prompt based on current_phase. Every prompt MUST include:

• Reference to the task file path (from queue's file field)
• The task identity (id, current_phase, target)
• The skill to invoke with --handoff
• ONE PHASE ONLY constraint
• Instruction to output RALPH HANDOFF block

Phase-specific prompts:

For extract phase (type=extract tasks only):

Read the task file at ops/queue/{FILE} for context.

You are processing task {ID} from the work queue.
Phase: extract | Target: {TARGET}

Run /reduce --handoff on the source file referenced in the task file.
After extraction: create per-claim task files, update the queue with new entries
(1 entry per claim with current_phase/completed_phases), output RALPH HANDOFF.
ONE PHASE ONLY. Do NOT run reflect or other phases.

For create phase:

Read the task file at ops/queue/{FILE} for context.

You are processing task {ID} from the work queue.
Phase: create | Target claim: {TARGET}

Create a {DOMAIN:note} for this claim in {DOMAIN:notes}/[claim as sentence].md
Follow note design patterns:
- YAML frontmatter with description (adds info beyond title), topics
- Body: 150-400 words showing reasoning with connective words
- Footer: Source (wiki link), Relevant Notes (with context), Topics
Update the task file's ## Create section.
ONE PHASE ONLY. Do NOT run reflect.

For enrich phase:

Read the task file at ops/queue/{FILE} for context.

You are processing task {ID} from the work queue.
Phase: enrich | Target: {TARGET}

Run /enrich --handoff using the task file for context.
The task file specifies which existing {DOMAIN:note} to enrich and what to add.
ONE PHASE ONLY. Do NOT run reflect.

For reflect phase:

Build sibling list: Query the queue for other claims in the same batch where completed_phases includes "create" (note already exists). Format as wiki links.

Read the task file at ops/queue/{FILE} for context.

You are processing task {ID} from the work queue.
Phase: reflect | Target: {TARGET}

OTHER CLAIMS FROM THIS BATCH (check connections to these alongside regular discovery):
{for each sibling in batch where completed_phases includes "create":}
- [[{SIBLING_TARGET}]]
{end for, or "None yet" if this is the first claim}

Run /reflect --handoff on: {TARGET}
Use dual discovery: {DOMAIN:topic map} exploration AND semantic search.
Add inline links where genuine connections exist — including sibling claims listed above.
Update relevant {DOMAIN:topic map} with this {DOMAIN:note}.
ONE PHASE ONLY. Do NOT run reweave.

For reweave phase:

Same sibling list as reflect (re-query queue for freshest state):

Read the task file at ops/queue/{FILE} for context.

You are processing task {ID} from the work queue.
Phase: reweave | Target: {TARGET}

OTHER CLAIMS FROM THIS BATCH:
{for each sibling in batch where completed_phases includes "create":}
- [[{SIBLING_TARGET}]]
{end for}

Run /reweave --handoff for: {TARGET}
This is the BACKWARD pass. Find OLDER {DOMAIN:note_plural} AND sibling claims
that should reference this {DOMAIN:note} but don't.
Add inline links FROM older {DOMAIN:note_plural} TO this {DOMAIN:note}.
ONE PHASE ONLY. Do NOT run verify.

For verify phase:

Read the task file at ops/queue/{FILE} for context.

You are processing task {ID} from the work queue.
Phase: verify | Target: {TARGET}

Run /verify --handoff on: {TARGET}
Combined verification: recite (cold-read prediction test), validate (schema check),
review (per-note health).
IMPORTANT: Recite runs FIRST — read only title+description, predict content,
THEN read full {DOMAIN:note}.
Final phase for this claim. ONE PHASE ONLY.

4c. Spawn Subagent (MANDATORY — NEVER SKIP)

Call the Task tool with the constructed prompt:

Task(
  prompt = {the constructed prompt from 4b},
  description = "{current_phase}: {short target}" (5 words max)
)

REPEAT: You MUST call the Task tool here. Do NOT execute the prompt yourself. Do NOT "optimize" by running the task inline. The Task tool call is the ONLY acceptable action at this step.

Wait for the subagent to complete and capture its return value.

4d. Evaluate Return

When the subagent returns:

1. Look for RALPH HANDOFF block — search for === RALPH HANDOFF and === END HANDOFF === markers
2. If handoff found: Parse the Work Done, Learnings, and Queue Updates sections
3. If handoff missing: Log a warning but continue — the work was still completed
4. Capture learnings: If Learnings section has non-NONE entries, note them for the final report

4e. Update Queue (Phase Progression)

After evaluating the return, advance the task to the next phase.

Phase progression logic:

Look up phase_order from the queue header to determine the next phase. Find current_phase in the array. If there is a next phase, advance. If it is the last phase, mark done.

If NOT the last phase — advance to next:

• Set current_phase to the next phase in the sequence
• Append the completed phase to completed_phases

If the last phase (verify) — mark task done:

• Set status: done
• Set completed to current UTC timestamp
• Set current_phase to null
• Append the completed phase to completed_phases

For extract tasks ONLY: Re-read the queue after marking done. The reduce skill writes new task entries (1 entry per claim/enrichment with current_phase/completed_phases) to the queue during execution. The lead must pick these up for subsequent iterations.

4f. Report Progress

=== Task {id} complete ({i}/{N}) ===
Phase: {current_phase} -> {next_phase or "done"}

If learnings were captured, show a brief summary. If more unblocked tasks exist, show the next one.

4g. Re-filter Tasks

Before the next iteration, re-read the queue and re-filter tasks. Phase advancement may have changed eligibility (e.g., after completing a create phase, the task is now at reflect — if filtering by --type reflect, it becomes eligible).

Step 5: Post-Batch Cross-Connect (Serial Mode)

After advancing a task to "done" (Step 4e), check if ALL tasks in that batch now have status: "done". If yes and the batch has 2 or more completed claims:

1. Collect all note paths from completed batch tasks. For each claim task with status: "done", read the task file's ## Create section to find the created note path.

2. Spawn ONE subagent for cross-connect validation:

Task(
  prompt = "You are running post-batch cross-connect validation for batch '{BATCH}'.

Notes created in this batch:
{list of ALL note titles + paths from completed batch tasks}

Verify sibling connections exist between batch notes. Add any that were missed
because sibling notes did not exist yet when the earlier claim's reflect ran.
Check backward link gaps. Output RALPH HANDOFF block when done.",
  description = "cross-connect: batch {BATCH}"
)

3. Parse handoff block, capture learnings. Include cross-connect results in the final report.

Skip if: batch has only 1 claim (no siblings) or tasks from the batch are still pending.