cd '/absolute/path/to/consumer-project' || exit 1
SF_PROJECT="$(pwd -P)"
cd '/absolute/path/to/trusted/software-factory' || exit 1
SF_PACK="$(pwd -P)"
SF_HOST='codex'; SF_PYTHON='python3'
"$SF_PYTHON" -I -B "$SF_PACK/install.py" install --host "$SF_HOST" --project "$SF_PROJECT" --profile full --skip-bmad --dry-run

Set up Software Factory in this Git project for Claude Code from this trusted local checkout. Read the pack README, AGENTS.md, installer help, and the project's existing instructions. Python 3.11+ and POSIX filesystem APIs are required; native Windows execution is unverified. Use physical paths (pwd -P), keep the pack and consumer distinct, and preserve user files and host settings.

From the pack checkout, preview the full project profile:
python3 -I -B install.py install --host claude-code --skip-bmad --project "<this-project>" --profile full --dry-run
Replace <this-project> with the physical consumer path. Inspect the destinations and conflicts; apply the same command without --dry-run only under the applicable local GO. This --skip-bmad route checks local project installation. Separately review and consent to the pinned official BMad 6.12.0 provisioning path before using it; observe actual BMad planner availability and structured epics/stories before claiming planning readiness. Missing capability is a named gap, not an invented plan.

Invoke /software-factory in the host UI. Use sf-help to read existing project context, prefill and reuse the twenty catalog decision slots in five groups, and ask only about material unknowns. Show the exact create-only setup proposal, digest, source decision and unknowns; reconcile the owner's setup GO before applying it. Preserve existing answers, documents and settings. Confirm actual host permissions, model/worker availability and any required Codex hook trust separately; installation alone grants none of these.

Ground one small real story in project facts and accepted BMad planning sources. Define its negative and positive eval cases, obtain actual consumer eval signoff before queuing covered work, and bind claims to the reviewed scope. Staff within host capacity, implement, independently verify and review the actual diff, repair failed or unknown proof, integrate within GO, and prove the resulting artifact effect before calling it delivered. Record the subject and real receipts. Continue the next eligible priority under finite delegated acceptance; escalate material risk, unclear authority, tradeoffs and reserved owner decisions at any phase. Keep optional bots and monitoring disabled until separately authorized; respect actual host permissions and never infer external access from local setup.

Set up Software Factory in this Git project for Cursor from this trusted local checkout. Read the pack README, AGENTS.md, installer help, and the project's existing instructions. Python 3.11+ and POSIX filesystem APIs are required; native Windows execution is unverified. Use physical paths (pwd -P), keep the pack and consumer distinct, and preserve user files and host settings.

From the pack checkout, preview the full project profile:
python3 -I -B install.py install --host cursor --skip-bmad --project "<this-project>" --profile full --dry-run
Replace <this-project> with the physical consumer path. Inspect the destinations and conflicts; apply the same command without --dry-run only under the applicable local GO. This --skip-bmad route checks local project installation. Separately review and consent to the pinned official BMad 6.12.0 provisioning path before using it; observe actual BMad planner availability and structured epics/stories before claiming planning readiness. Missing capability is a named gap, not an invented plan.

Invoke /software-factory in the host UI. Use sf-help to read existing project context, prefill and reuse the twenty catalog decision slots in five groups, and ask only about material unknowns. Show the exact create-only setup proposal, digest, source decision and unknowns; reconcile the owner's setup GO before applying it. Preserve existing answers, documents and settings. Confirm actual host permissions, model/worker availability and any required Codex hook trust separately; installation alone grants none of these.

Ground one small real story in project facts and accepted BMad planning sources. Define its negative and positive eval cases, obtain actual consumer eval signoff before queuing covered work, and bind claims to the reviewed scope. Staff within host capacity, implement, independently verify and review the actual diff, repair failed or unknown proof, integrate within GO, and prove the resulting artifact effect before calling it delivered. Record the subject and real receipts. Continue the next eligible priority under finite delegated acceptance; escalate material risk, unclear authority, tradeoffs and reserved owner decisions at any phase. Keep optional bots and monitoring disabled until separately authorized; respect actual host permissions and never infer external access from local setup.

Set up Software Factory in this Git project for Codex from this trusted local checkout. Read the pack README, AGENTS.md, installer help, and the project's existing instructions. Python 3.11+ and POSIX filesystem APIs are required; native Windows execution is unverified. Use physical paths (pwd -P), keep the pack and consumer distinct, and preserve user files and host settings.

From the pack checkout, preview the full project profile:
python3 -I -B install.py install --host codex --skip-bmad --project "<this-project>" --profile full --dry-run
Replace <this-project> with the physical consumer path. Inspect the destinations and conflicts; apply the same command without --dry-run only under the applicable local GO. This --skip-bmad route checks local project installation. Separately review and consent to the pinned official BMad 6.12.0 provisioning path before using it; observe actual BMad planner availability and structured epics/stories before claiming planning readiness. Missing capability is a named gap, not an invented plan.

Invoke $software-factory in the host UI. Use sf-help to read existing project context, prefill and reuse the twenty catalog decision slots in five groups, and ask only about material unknowns. Show the exact create-only setup proposal, digest, source decision and unknowns; reconcile the owner's setup GO before applying it. Preserve existing answers, documents and settings. Confirm actual host permissions, model/worker availability and any required Codex hook trust separately; installation alone grants none of these.

Ground one small real story in project facts and accepted BMad planning sources. Define its negative and positive eval cases, obtain actual consumer eval signoff before queuing covered work, and bind claims to the reviewed scope. Staff within host capacity, implement, independently verify and review the actual diff, repair failed or unknown proof, integrate within GO, and prove the resulting artifact effect before calling it delivered. Record the subject and real receipts. Continue the next eligible priority under finite delegated acceptance; escalate material risk, unclear authority, tradeoffs and reserved owner decisions at any phase. Keep optional bots and monitoring disabled until separately authorized; respect actual host permissions and never infer external access from local setup.

flowchart LR
  A[Ground intent and existing facts] --> B[Plan with BMad skills and review under scoped GO]
  B --> C[Host staffs claimed work]
  C --> D[Implement and independently verify]
  D -->|failed or unknown| C
  D --> E[One owner integrates and checks combined SHA]
  E --> F[Prove affected behavior on resulting artifact]
  F -->|failed or unknown| C
  F --> G{Material risk or reserved owner decision?}
  G -->|yes| I[Owner decides within actual scope]
  I --> H[Record acceptance provenance and next eligible priority]
  G -->|no| H
  H --> A

flowchart TD
  A[Ground repo, principles and desired outcome] --> B[Plan with observed BMad skills and structured stories]
  B --> X{Planning capability and source observed?}
  X -->|missing| Y[Record readiness gap; reconcile scoped alternative]
  Y -->|approved equivalent path| C[Independent plan review]
  X -->|yes| C
  C -->|material finding| B
  C --> D{Applicable GO and required decisions?}
  D -->|missing material decision| E[Ask its owner; block dependents only]
  E --> D
  D -->|yes| F[Select eligible frontier; claim exact paths]
  F --> G[Worker implements, tests and self-reviews]
  G --> H[Independent QA on actual HEAD and eval cases]
  H -->|failed or unknown| G
  H --> I[Actual-diff review and applicable CI or bot proof]
  I -->|confirmed finding| G
  I --> J[Check in review receipts while claim is held]
  J --> K[Reconcile handoff; release claim]
  K --> L[One integration owner; providers done; strict gate]
  L -->|repair needed| F
  L --> M[Integrate within GO; check combined SHA]
  M --> N[Authorized staging or approved local-equivalent effect proof]
  N -->|failed or unknown| F
  N --> O[Record story and resulting-artifact evidence]
  O --> P{Material risk or reserved owner decision?}
  P -->|yes| Q[Owner decision with source and scope]
  P -->|no| R[Evidence-bound acceptance and next eligible work]
  Q --> R

flowchart TD
  A[Report, failed eval or authorized signal] --> B[Durable bug identity and primary evidence]
  B --> C[Safe reproduction; mechanism distinct from hypothesis]
  C --> D{What does evidence support?}
  D -->|unknown| E[Separating experiment or targeted clarification]
  E --> B
  D -->|correct or already handled| U[Independent reproduction and disposition review]
  U -->|supported| F[No-fix investigative closure; zero code change]
  U -->|unconfirmed or disputed| B
  F --> G[Retain reproduction and disposition]
  D -->|confirmed in-scope defect| H[Risk-sized plan and regression or eval cases]
  H --> I[Independent plan review; claim work]
  I --> J[Implement, test and independently verify actual HEAD]
  J -->|failed or unknown| I
  J --> K[Actual-diff review and applicable checks]
  K -->|confirmed finding| I
  K --> L[Single-owner gated integration under GO]
  L --> M[Original reproduction on resulting artifact]
  M -->|failed or unknown| I
  M --> N[Durable closure; assess applicable acceptance and escalation]
  G --> O[Authorized re-observation within source and window]
  N --> O
  O -->|recurs after closure| P[Linked follow-up; preserve terminal record]
  O -->|new authority needed| Q[Owner decision; independent approved work continues]

flowchart TD
  A[Trust pack source; choose physical consumer root and host] --> B[Choose project profile and preview plan]
  B --> C{Full or skills-only?}
  C -->|full| D[Skills, roles, policies, hooks and auxiliary assets]
  C -->|skills-only| E[Skills and required auxiliary assets; reduced integration]
  D --> F[Inspect ownership, collisions and preserved user bytes]
  E --> F
  F -->|unsafe or conflicting| G[Refuse; preserve bytes; owner reconciles]
  F -->|reviewed local install scope| H[Installer lock and source revalidation]
  H --> I[Conditional publication and ownership manifest]
  I -->|concurrent edit or source change| G
  I --> J[Inspect resulting files; repeat is unchanged]
  J --> K{Separately consented BMad provisioning?}
  K -->|no, skip-bmad| L[Installer smoke only; BMad planning remains unverified]
  K -->|yes| M[Pinned official installer; prerequisites and output checks]
  M -->|missing or failed| N[Record planning-readiness gap or partial state; no implicit retry]
  M --> O[Observe installed modules; workflow quality still unproved]
  J --> P[Doctor: report checks and unknowns; gh auth may run]
  P --> Q[Dry-run uninstall; selected ownership only]
  Q --> R[Authorized uninstall preserves edits and user context]

flowchart LR
  A[Eleven canonical skills and role prompts] --> B[Installer or allowlisted native bundle generator]
  C[Canonical host policy and required auxiliary assets] --> B
  B --> D[Claude Code generated layout]
  B --> E[Cursor generated layout]
  B --> F[Codex generated layout]
  D --> G[Active host loads supported instructions]
  E --> G
  F --> G
  G --> H[Host orchestrator chooses eligible work and staff]
  H --> I[Actual host tools, permissions and available models]
  I --> J[Workers act only within applicable GO]
  J --> K[Local helpers validate, record or project bounded facts]
  K --> H
  L[Read-only registry and routing proposals] --> H
  M[Static format checks] --> N[Format evidence only; live loading remains separate]
  B --> M

flowchart TD
  A[Inspect consumer intent and existing documents] --> B{Existing repo without grounded crawl?}
  B -->|yes| C[Bounded read-only crawl; save derived map]
  B -->|no| D[Read-only setup discovery]
  C --> D
  D --> E[Twenty catalog slots; reuse confirmed facts and answers]
  E --> F{Material unknown remains?}
  F -->|yes| G[One targeted owner decision; keep independent work moving]
  G --> E
  F -->|no| H[One action for safe defaults; propose exact missing drafts]
  H --> I[Inspect config, document paths, unknowns and source digest]
  I --> J[Host reconciles actual GO and approval source]
  J --> K[Apply exact approved proposal; create only]
  K -->|stale, collision, unsafe root or partial state| L[Refuse or preserve partial bytes; report exact next action]
  K --> M[Validate actual config and retained documents]
  M --> N[Re-onboard reuses answers; no new approval claim]
  M --> O[Real consumer eval signoff before queue]
  O --> P[Host continues scoped delivery; optional access stays unknown]

flowchart TD
  A[Observed rework or bounded improvement opportunity] --> B[Read principles, source revision and applicable GO]
  B --> C{Smallest useful path?}
  C -->|capability| D[Inspect local metadata, license, digest and permission needs]
  D -->|unknown| E[Defer adoption; native work can continue]
  C -->|BMad| F[Inspect actual pinned customization and callback recursion]
  C -->|Markdown customization| G[Capture exact base, payload and evidence]
  F --> H[Bounded proposal; protected governing policy stays owned]
  G --> H
  D -->|adequate facts| H
  H --> I[Independent review plus regression and canary artifacts]
  I -->|failed or unknown| H
  I --> J{Standing GO sufficient?}
  J -->|material trust, policy or external effect| K[Human source decision]
  K --> L[Host reconciles real authority and evidence]
  J -->|routine and covered| L
  L --> R{Approved change kind?}
  R -->|Markdown overlay| M[Installer owns overlay composition and publication]
  R -->|capability or MCP| S[Actual scoped host or tool adoption; registry never installs or connects]
  S -->|unavailable or unproved| E
  R -->|BMad customization| T[Host applies scoped customization through its actual supported method]
  M -->|drift or edited target| H
  M --> N[Inspect actual result and applicable ownership; observe authorized later run]
  S -->|actually adopted| N
  T --> N
  N --> O{Affected behavior improved?}
  O -->|yes| P[Retain with evidence and next retro]
  O -->|no or unknown| Q[Propose scoped rollback; prove restored behavior]
  Q --> H

flowchart TD
  A[Ground requirements, risk and applicable surfaces] --> B[Stable fail-to-pass and pass-to-pass cases]
  B --> C[Owner-approved plan and exact signoff payload]
  C --> D[Queue only covered applicable work]
  D --> E[Run actual cases on identified subject]
  E --> F{Harness and required proof available?}
  F -->|no| G[Unverified or blocked; preserve missing prerequisite]
  F -->|yes| H[Retain actual output, artifacts and verdicts]
  H --> I{Pass with intact current-subject receipts?}
  I -->|failed, stale, tampered or unknown| J[Triage and repair; preserve prior verdicts]
  G --> J
  J --> E
  I -->|yes| K[Independent evidence review and applicable strict board gate]
  K -->|not adequate| J
  K --> L[Integrate; refresh affected proof on combined SHA]
  L --> M[Prove resulting artifact effect separately]
  J --> N[Exceptional override needs actual scoped owner decision]
  N --> O[Preserve cases and original verdict; record reason and source]
  O --> K

flowchart TD
  A[Planning source with stable source keys] --> B[Owner derives immutable story and epic identities]
  B --> C[Per-story records are the only live story-status authority]
  C --> D[Readiness from dependency edges and applicable accepted epic gates]
  D --> E[In-review, built or done providers may unlock start; explicit order ranks]
  E --> F[Claim exact paths; worker sets in-progress]
  F --> G[Receipts and in-review while claim is held]
  G --> H[Reconcile handoff; release ownership]
  H --> I[All providers done and strict premerge gate to built]
  I --> J[Actual merge and merge-subject evidence to done]
  J --> K[All children done or cancelled: awaiting acceptance]
  K --> L[Evidence-bound delegated or reserved human epic decision]
  C --> M[Owner regenerates roadmap, dashboard and graph]
  N[Kernel claims and checkpoints: execution and recovery facts] --> M
  M --> O[Timestamped derived views and read-only graph questions]
  O --> P[Recorded attribution is not proof of live workers or effects]
  C --> Q[Blocked overlay retains status and exact next action]
  Q --> D
  C --> R[Scoped confirmed cancellation is terminal]
  R --> S[Cancelled prerequisites still block until explicit reconciliation]