plan-migration-partition

name	plan-migration-partition
description	Use at M2 PLAN — before minting Kanban children — to follow the Hermes skills speckit-specify, speckit-plan, and speckit-tasks (never implement), read M1 attachments, author evidence/partition.json, convert with k4_convert.py --partition --tasks, and mint with k4_mint.py. Use when writing partition.json, even if the user does not name Spec Kit. Do not pin this leaf on the card (pin paved-road-m2). Do not run specify workflow run speckit: the Spec Kit hermes integration installs files:{} and cannot dispatch speckit-specify as a command. Do not PATH-lookup specify. Do not scrape write-sets from tasks.md. Do not use only to lint an already-written partition (check-spec-readiness).
license	Apache-2.0
compatibility	Linux seat; Python 3.11+; pinned specify-cli; Hermes Kanban
metadata	{"author":"rhoai3-harness-team","version":"1.5.0","hermes":{"tags":["sdd","m2"],"category":"sdd","kind":"guidance"}}

Plan the migration partition (M2 producer)

This skill is the M2 producer of evidence/partition.json. check-spec-readiness defers to it. Do not pin this leaf on the card. Pin --skill paved-road-m2; this producer loads via skill_view from steps.json. Speckit skills generate the plan (tasks.md); K4 generates cards from the typed partition (Architect 142518ZA). Do not grep tasks.md for write-sets (PATH_TOKEN OBJECT). Do not specify workflow run speckit (Architect 170540ZA: hermes.manifest files: {}).

When to Use

• This card is M2 PLAN.
• evidence/partition.json does not exist yet, or M1 attachments changed.
• Not Spec Kit provision (init-spec-workspace).
• Not coverage/body lint alone (check-spec-readiness).
• Not M3 exit derivation (derive-story-oracles).

Card pin

Do not pin this leaf on the M2 card. Dest-init autostart-migration.sh pins --skill paved-road-m2 only. Follow this SKILL.md when skill_view loads it from steps.json. init-spec-workspace and derive-story-oracles are not the procedure.

Procedure

Root is /projects/modernized. $M1 is the parent M1 task id.

1. If .specify/ is missing, load init-spec-workspace and provision. Then return here.
2. Read M1 attachments — not metadata path lists:

hermes kanban show "$M1"
# then read kanban_attachments under $HERMES_HOME/kanban/attachments/<m1-id>/
# and evidence/findings-handoff.json, entry-point-inventory.json,
# type-inventory.json, required-extensions.json

3. Follow the Hermes speckit skills in process. Stop. Never /speckit.implement. Never specify workflow run speckit — Spec Kit 0.16.1 --integration hermes writes .specify/integrations/hermes.manifest.json with "files": {}. invoke_separator: "-" then dispatches speckit.specify as the shell command speckit-specify, which does not exist (Unknown skill(s): speckit-specify on dest-9/10/12). The SKILL.md files seed-speckit-skills.py installs are for this agent to read, not commands a workflow runner can exec (Architect 170540ZA). Derive the spec from M1 attachments (identity, HTTP paths, extensions) — do not invent routes.

   Load and follow, in order, the project leaves:

   • .hermes/skills/speckit-specify/SKILL.md
   • .hermes/skills/speckit-plan/SKILL.md
   • .hermes/skills/speckit-tasks/SKILL.md

   Do not load speckit-implement. Do not PATH-lookup specify. Do not widen K2_ALLOW_ROOT. Do not dest-edit dest-9 PATH or implementer external_dirs. If a named skill is missing, a named command fails, or a named path is absent: stop. Record exactly what failed. kanban_block. Do not hand-author tasks.md. Do not stamp evidence/receipts/speckit/workflow-run.json to satisfy a gate (Architect 170112ZA: that receipt is worker-writable).

tasks.md must be non-empty at the Spec Kit 0.16.1 feature directory named in .specify/feature.json feature_directory (example: specs/001-spring-to-quarkus/tasks.md). Do not copy that file onto .specify/specs/ to satisfy a glob. Do not glob .specify/specs/*/tasks.md as the M2 authority.

4. Author evidence/partition.json from those attachments and the plan. Schema: references/partition-schema.md (kept in sync with .hermes/kernel/k4_schema.py and .hermes/kernel/k4_producers.py by scripts/assert-partition-schema-sync.py). Pin skills[] from that page's Valid producers table (or omit and use kind → defaults). Do not reverse-engineer k4_*.py. Do not invent skill names.

   Write-set (K4_T0_3_SERVICE): the same *Service.java must not appear on two stories. One story MAY own a shared facade. Split-one-class-per-aggregate is petclinic architecture, not this mint refuse.

   HTTP stories must set legacy_source to the matching entry-point-inventory.json file (legacy package/path). Dest files_writable is the dest package (com/demo); the worker must not rediscover com/example/restservice. Missing legacy_source is K4_LEGACY_SOURCE. HTTP stories must also set dest_file to the type-inventory dest twin of an inventoried legacy type — not the new file this story creates. Missing dest_file is K4_DEST_FILE (dest-9 live partition skip is convert-refuse, not a second invented-files checker).

5. Story headings in tasks.md must match partition story_ids (scripts/assert-m2-story-headings.py). Still not write-set scrape.

6. Convert, then mint:

python3 .hermes/kernel/k4_convert.py \
  --partition evidence/partition.json \
  --tasks specs/*/tasks.md \
  --out evidence/partition-payloads.json
python3 .hermes/kernel/k4_mint.py \
  --payloads evidence/partition-payloads.json \
  --exec

--tasks is required for a conformant M2 complete (Operator 123401ZO). A hand-written partition remains a legal K4 input; it is not a complete. k4_mint.py --exec mints the M3 stories, then one M4 VERIFY parented to those M3 t_* (--idempotency-key m4-verify). Do not extend autostart-migration.sh to mint M4. Do not put a verdict token in the M4 body.

7. Lint (this skill does not replace these checkers):

python3 .hermes/skills/sdd/check-spec-readiness/scripts/assert-m2-speckit-conformance.py .
python3 .hermes/skills/sdd/check-spec-readiness/scripts/check-partition-coverage.py .
bash .hermes/skills/sdd/check-spec-readiness/scripts/check-readiness.sh --root .
python3 "${HERMES_SKILL_DIR}/scripts/assert-m2-story-headings.py" .

Happy-path terminator is kanban_request_review (not kanban_complete). --metadata created_cards is the native t_* list from mint. KEEP evidence/partition-payloads.json and the official log also prove mint. Empty after a mint is OBJECT. Do not kanban daemon --force. Do not kanban swarm. Do not dest-dispatch M5.

Pitfalls

• Assembling the sequence from three other skills plus AGENTS.md (dest-8 restated the protocol six times and read k4_*.py).
• Skipping speckit because K4 does not scrape tasks.md (WAVE 1 made tasks.md binding, not generative of cards).
• Inventing /q/health as a story (constitution III: if health exists, at /q/health; VII stands).
• Prescribing specify workflow run speckit (Operator 165811ZO bare form; Architect 170540ZA hermes files: {}). Follow speckit-specify / speckit-plan / speckit-tasks as Hermes skills. dest-9/10/12 Unknown skill(s): speckit-specify is kanban_block, not a hand-authored tasks.md and not a stamped workflow-run receipt.