Flatten the skill chain: short dispatchers, plan_branches folded, mandatory-load budget

bin/README.md

@@ -18,7 +18,7 @@ Auto-discovers (no hardcoded skill list) by scanning each organ repo's
 
 - **`PyAutoMind/skills/`** — registry-coupled skills (`create_issue`).
 - **`PyAutoBrain/skills/`** — development-workflow skills (`start_dev`,
-  `start_dev_for_user`, `plan_branches`, `start_library`, `start_workspace`,
+  `start_dev_for_user`, `start_library`, `start_workspace`,
   `ship_library`, `ship_workspace`, `register_and_iterate`, `repo_cleanup`,
   `update_issue`).
 - **`PyAutoHeart/skills/`** — status / readiness / validation skills

bin/check_skill_line_counts.sh

@@ -35,25 +35,31 @@ for root in "${ROOTS[@]}"; do
   for dir in "$root"/*/; do
     [ -d "$dir" ] || continue
     name="$(basename "$dir")"
-    # Primary files: SKILL.md (dispatcher/frontmatter) and the command/body
-    # <dirname>.md. Supporting docs (reference.md, examples, etc.) are exempt.
+    # The budget is the skill's MANDATORY LOAD: SKILL.md (dispatcher/
+    # frontmatter) plus the command/body <dirname>.md, summed — both are read
+    # on every invocation. Supporting docs (reference.md, examples, etc.) are
+    # lazy and exempt.
+    total=0
+    found=0
     for primary in "$dir/SKILL.md" "$dir/$name.md"; do
       [ -f "$primary" ] || continue
-      checked=$((checked + 1))
-      lines=$(wc -l < "$primary")
-      if [ "$lines" -gt "$LIMIT" ]; then
-        echo "OVER  ${lines}  ${primary#$PYAUTO_ROOT/}"
-        over=$((over + 1))
-      fi
+      found=1
+      total=$((total + $(wc -l < "$primary")))
     done
+    [ "$found" -eq 1 ] || continue
+    checked=$((checked + 1))
+    if [ "$total" -gt "$LIMIT" ]; then
+      echo "OVER  ${total}  ${dir#$PYAUTO_ROOT/} (SKILL.md + $name.md)"
+      over=$((over + 1))
+    fi
   done
 done
 
 echo ""
 if [ "$over" -eq 0 ]; then
-  echo "OK: all $checked primary skill files are within $LIMIT lines."
+  echo "OK: all $checked skills are within the $LIMIT-line mandatory-load budget."
   exit 0
 else
-  echo "FAIL: $over primary skill file(s) exceed $LIMIT lines — factor detail into reference.md."
+  echo "FAIL: $over skill(s) exceed the $LIMIT-line mandatory-load budget — factor detail into reference.md."
   exit 1
 fi

bin/install.sh

@@ -11,7 +11,7 @@
 #                             slated to leave PyAutoLabs/; kept only so an old
 #                             checkout still resolves, auto-skipped once gone
 #   - PyAutoMind/skills/    — registry-coupled skills (create_issue, handoff)
-#   - PyAutoBrain/skills/   — development-workflow skills (start_*/ship_*/plan_branches/…)
+#   - PyAutoBrain/skills/   — development-workflow skills (start_*/ship_*/…)
 #   - PyAutoHeart/skills/   — status / readiness / diagnostic skills
 #   - PyAutoBuild/skills/   — release-execution skills ONLY (pre_build)
 #   - autolens_profiling/skills/ — science-profiling skills (profile_likelihood)

skills/WORKFLOW.md

@@ -1,6 +1,6 @@
 # PyAuto workflow skills — shared reference
 
-The `start_*` / `ship_*` / `plan_branches` / `register_and_iterate` skills are the
+The `start_*` / `ship_*` / `register_and_iterate` skills are the
 **development-workflow entry points** of the PyAuto organism. They are *not*
 independent reasoning systems: each one is a thin entry point that delegates to
 the organs. This file is the shared context every workflow skill points at, so
@@ -133,6 +133,19 @@ pyauto-heart readiness --json    # authoritative GREEN / YELLOW / RED verdict
 Tests/smoke runs that feed the verdict are Heart's domain — invoke them through
 the vitals faculty rather than re-deriving pass/fail criteria in the skill.
 
+## Cross-harness notes (apply to every workflow skill)
+
+- `/name` references mean "use that skill"; a harness without slash commands
+  follows the same body file directly.
+- "Plan Mode" means: present the plan and wait for explicit user approval
+  before any file edit.
+- If the user gives a development task with **no** PyAutoMind prompt path,
+  first write a concise prompt under the right `<work-type>/<target>/` folder
+  (original request verbatim), then continue with that path.
+- Where a body delegates mechanical execution to a Sonnet subagent, a harness
+  without subagents performs the same steps directly, preserving the
+  judgment/mechanical split above.
+
 ## Execution environments
 
 There is **no special "mobile" or "phone" mode** — PyAutoBrain runs the same

skills/repo_cleanup/SKILL.md

@@ -30,8 +30,8 @@ execution-environment model: [`../WORKFLOW.md`](../WORKFLOW.md).
 
 **Distinct from:** `worktree_status` (Heart read-only diagnostic — consulted here,
 but this also mutates); post-merge cleanup in `CLAUDE.md` (once per shipped task —
-this covers residue when that flow is skipped); `plan_branches` (task start — this
-is between-tasks hygiene).
+this covers residue when that flow is skipped); the start_dev branch survey
+(task start — this is between-tasks hygiene).
 
 ## Safety principles (non-negotiable)
 

skills/ship_library/SKILL.md

@@ -5,25 +5,11 @@ description: Ship PyAutoLabs source-library changes — run tests, commit, push,
 
 # Ship Library
 
-A PyAutoBrain development-workflow entry point. Shipping a library feature is
-**feature-dev** work: dev-workflow → vitals faculty → Heart readiness gate →
-commit/push/feature-PR, with PyAutoMind holding task state. It is **not** a Build
-task — Build is release/packaging only (PyPI/tags/notebooks) and `ship_*` reaches
-it solely for the release step. Shared organ boundary, the readiness gate and the
-execution-environment model are in [`../WORKFLOW.md`](../WORKFLOW.md).
-
-Use `ship_library.md` in this directory as the authoritative workflow body
-(`reference.md` holds the PR format, execution contract and impact analysis).
-
-Follow the command file exactly, adapting Claude-specific references to Codex:
-
-- `/ship_library` means use this skill.
-- If the command file delegates mechanical execution to a Claude subagent, Codex
-  should either use an available subagent tool with the same contract or perform
-  the same mechanical steps directly while preserving the judgment/mechanical
-  split in the user-facing workflow.
-- Preserve the `## API Changes` PR-body contract because `/start_workspace`,
-  release review, and downstream workspace migration depend on it.
-
-Do not duplicate or reinterpret the workflow here. If the workflow changes, edit
-`ship_library.md`.
+Follow [`ship_library.md`](ship_library.md) in this directory exactly — the
+authoritative workflow body (`reference.md` holds the PR format, execution
+contract and impact analysis). Shared context and cross-harness notes:
+[`../WORKFLOW.md`](../WORKFLOW.md). Shipping a library feature is
+**feature-dev** work gated by Heart — not a Build task. Preserve the
+`## API Changes` PR-body contract — `/start_workspace`, release review, and
+downstream workspace migration depend on it. Do not duplicate or reinterpret
+the workflow here — if it changes, edit `ship_library.md`.

skills/ship_workspace/SKILL.md

@@ -5,26 +5,11 @@ description: Ship PyAutoLabs workspace and tutorial changes — validate scripts
 
 # Ship Workspace
 
-A PyAutoBrain development-workflow entry point. Shipping a workspace feature is
-**feature-dev** work: dev-workflow → vitals faculty → Heart readiness gate →
-commit/push/feature-PR/merge, with PyAutoMind holding task state. It is **not** a
-Build task — Build is release/packaging only. Shared organ boundary, the
-readiness gate and the execution-environment model are in
-[`../WORKFLOW.md`](../WORKFLOW.md).
-
-Use `ship_workspace.md` in this directory as the authoritative workflow body
-(`reference.md` holds the PR format, merge gate and issue/Mind formats).
-
-Follow the command file exactly, adapting Claude-specific references to Codex:
-
-- `/ship_workspace` means use this skill.
-- Slash-command references such as `/smoke_test` refer to the matching Codex
-  skill or shared command body.
-- If the command file delegates mechanical execution to a Claude subagent, Codex
-  should either use an available subagent tool with the same contract or perform
-  the same mechanical steps directly.
-- Preserve the `## Scripts Changed` PR-body contract and the library-first merge
-  gate for linked workspace PRs.
-
-Do not duplicate or reinterpret the workflow here. If the workflow changes, edit
-`ship_workspace.md`.
+Follow [`ship_workspace.md`](ship_workspace.md) in this directory exactly — the
+authoritative workflow body (`reference.md` holds the PR format, merge gate and
+issue/Mind formats). Shared context and cross-harness notes:
+[`../WORKFLOW.md`](../WORKFLOW.md). Shipping a workspace feature is
+**feature-dev** work gated by Heart — not a Build task. Preserve the
+`## Scripts Changed` PR-body contract and the **library-first merge gate** for
+linked workspace PRs. Do not duplicate or reinterpret the workflow here — if it
+changes, edit `ship_workspace.md`.

skills/start_dev/SKILL.md

@@ -5,25 +5,8 @@ description: Default entry point for PyAutoLabs development tasks — implement,
 
 # Start Dev
 
-A PyAutoBrain development-workflow entry point. It routes reasoning through the
-Brain Feature Agent, consults PyAutoMemory, records state in PyAutoMind, and
-hands off to `/start_library` / `/start_workspace`. Shared organ boundary,
-agent entry points and the execution-environment model are in
-[`../WORKFLOW.md`](../WORKFLOW.md).
-
-Use `start_dev.md` in this directory as the authoritative workflow body
-(`reference.md` holds the long-form templates).
-
-Follow the command file exactly, adapting Claude-specific references to Codex:
-
-- `/start_dev <prompt>` means use this skill with the same prompt path.
-- If the user gives a development task without a prompt path, first create a
-  concise prompt file in the appropriate `PyAutoMind/` category and include
-  the original request verbatim, then use this skill with that prompt path.
-- "Plan Mode" means present the plan and wait for explicit user approval before
-  file edits when this workflow is being used.
-- Slash-command references such as `/plan_branches`, `/start_library`, and
-  `/start_workspace` refer to the matching Codex skill or shared command body.
-
-Do not duplicate or reinterpret the workflow here. If the workflow changes, edit
-`start_dev.md`.
+Follow [`start_dev.md`](start_dev.md) in this directory exactly — the
+authoritative workflow body. `reference.md` holds the long-form templates and
+the branch survey; shared context and cross-harness notes are in
+[`../WORKFLOW.md`](../WORKFLOW.md). Do not duplicate or reinterpret the
+workflow here — if it changes, edit `start_dev.md`.