[Preset]: Update Autonomous Run Governance to v0.4.4

### Preset ID

autonomous-run-governance

### Preset Name

Autonomous Run Governance

### Version

0.4.4

### Description

Adds permission-bounded autonomous delivery with index-bound staged paths, regular-file modes and bytes, semantic completion, exact-head evidence, and hash-bound historical whitespace exceptions.

### Author

Thorsten Hindermann

### Repository URL

https://github.com/hindermath/spec-kit-preset-autonomous-run-governance

### Download URL

https://github.com/hindermath/spec-kit-preset-autonomous-run-governance/archive/refs/tags/v0.4.4.zip

### Documentation URL

https://github.com/hindermath/spec-kit-preset-autonomous-run-governance/blob/v0.4.4/README.md

### License

MIT

### Required Spec Kit Version

>=0.8.3

### Required Extensions (optional)

None

### Templates Provided

- `autonomous-run-governance-model-routing` - provider-neutral command-to-role composition contract
- `constitution-template` - autonomous delivery authority and safety governance
- `spec-template` - autonomous-run applicability and scope boundaries
- `plan-template` - convergence, evidence, validation, and closeout planning
- `tasks-template` - dependency-ordered autonomous delivery tasks
- `autonomous-run-agent-guidance-addendum-template` - portable agent guidance
- `autonomous-runbook-template` - permission-bounded execution runbook
- `autonomous-run-evidence-template` - convergence, validation, and delivery ledger
- `autonomous-run-state-template` - machine-readable lifecycle checkpoint
- `autonomous-runner-profiles-template` - local fail-closed model bindings
- `autonomous-run-gate-requirements-template` - declared acceptance gates
- `autonomous-run-gate-evidence-template` - exact-head gate evidence
- `autonomous-phase-result-template` - semantic phase completion result
- `autonomous-run-retrospective-template` - classified workflow learning
- `autonomous-run-readiness-checklist-template` - delivery readiness
- `autonomous-run-field-validation-summary` - field-validation summary

### Commands Provided

- `speckit.autonomous` - executes one complete Spec Kit feature under explicit delivery authority
- `speckit.autonomous-status` - inspects an active or interrupted run without mutation
- `speckit.autonomous-stop` - cooperatively pauses at a safe boundary
- `speckit.autonomous-resume` - resumes after drift, operation, governance, and authority revalidation
- `speckit.autonomous-retrospective` - extracts and classifies reusable learning

### Number of Scripts (optional)

11

### Tags

autonomous, governance, evidence, permissions, sdd

### Key Features

This v0.4.4 submission supersedes the unaccepted v0.4.3 update in this same issue. No second preset submission was opened. The community catalog still references v0.4.1.

This is an update of the existing `autonomous-run-governance` catalog entry from v0.4.1 to v0.4.4, not a new preset ID. The previous submission is #4153; the live community catalog still references v0.4.1 at submission time.

- Corrected in v0.4.4: staged path inventory is read without rename collapsing, so both the deleted source and added target must be explicitly intended. File modes and bytes are read from the Git index; only regular stage-0 entries are accepted, so staged symlinks, gitlinks, conflicts, and worktree/index type races fail closed.
- Preserved from v0.4.3: narrowly scoped historical trailing-whitespace allowances apply only to genuinely untracked additions and remain valid through exact staged-candidate validation. Each allowance binds one exact repository-relative path and the SHA-256 of its original bytes.
- Explicit current approval is required by the workflow contract. The deterministic validator rejects malformed, duplicate, mismatched, unused, or unnecessary allowances. Content drift invalidates the hash. The default trailing-whitespace rejection remains in effect for every other file.
- Both Bash (`--allow-historical-whitespace PATH=RAW_SHA256`) and PowerShell (`-AllowHistoricalWhitespace PATH=RAW_SHA256`) expose the same contract. Successful JSON evidence records the path/hash allowance; index and worktree immutability checks remain in place.
- Unchanged: semantic phase completion beyond exit code zero, lifecycle-bound exact-head PreMerge/PostMerge evidence, explicit delivery modes, read-only status, cooperative stop, revalidated resume, optional intake review, and fail-closed model routing. Intake Review Governance is an optional companion preset activated by repository policy, not a required extension.
- The provider-neutral `model-routing.json` composition contract is preserved; concrete model names remain local. Installation grants no commit, push, PR, merge, bypass, secret, or administrative authority.
- Counting clarification: the manifest declares 16 `type: template` contributions (15 workflow templates plus the existing model-routing contract), five commands, and 11 scripts. The routing contribution was already present in v0.4.1; this is not an additional template introduced by the patch.

Release and verification:

- Release: https://github.com/hindermath/spec-kit-preset-autonomous-run-governance/releases/tag/v0.4.4 (regular release, not pre-release).
- Source correction and regression proof: https://github.com/hindermath/spec-kit-preset-autonomous-run-governance/pull/20. This records Bash and PowerShell execution against three unchanged Feature-032 reports. Earlier independent real-project evidence remains documented in the versioned field-validation summary; no new full product run is claimed here.
- Annotated tag v0.4.4 resolves to commit `f718c00d405260706709be2fd64f59aadfac9312`, matching main when checked.
- Downloaded tag-ZIP SHA-256: `2478d5fe193abe677601a024f5fe18efae635e8a94bba66b50b59c4b34a8fd8d`.
- Fresh archive installation, preset list/info, and resolution of state, phase-result, runbook, and model-routing templates were rechecked on macOS on 2026-09-12 with Specify CLI 0.12.8. Command generation is checked separately in an initialized Claude integration; `preset resolve` is for templates, not command entries. The installed package matches the downloaded archive.
- `bash tests/test-autonomous-evidence-integrity.sh` executed from the unpacked v0.4.4 archive: PASS, exit 0, including Bash/PowerShell rename-inventory, index-only symlink, gitlink, historical-whitespace, lifecycle, and semantic-evidence regressions.
- `bash tests/test-model-routing.sh` executed from the tag-identical source checkout within the existing twelve-preset composition: PASS, exit 0. This test intentionally requires all twelve neighboring routing catalogs; the isolated single-archive attempt blocked on the absent `speckit.intake-read` companion contract. No package or test was modified to obtain the passing composition result.
- The initialized Claude integration generated all five autonomous skills. No fresh Linux/Windows-host run or new full product run is claimed by this submission; fleet rollout validation is tracked separately.

### Testing Checklist

- [x] Preset installs successfully via `specify preset add`
- [x] Template resolution works correctly after installation
- [x] Documentation is complete and accurate
- [x] Tested on at least one real project

### Submission Requirements

- [x] Valid `preset.yml` manifest included
- [x] Linked README (Documentation URL) explains how to use this preset and includes a valid `specify preset add ...` command using the exact Download URL
- [x] LICENSE file included
- [x] GitHub release created with version tag
- [x] Preset ID follows naming conventions (lowercase-with-hyphens)

Start by comparing the existing autonomous-run-governance catalog entry with submission #4153 and the v0.4.4 release manifest, README, and download archive. Verify the preset metadata, version, templates, commands, and installation details, then run the listed evidence-integrity and model-routing tests. Done means the catalog references v0.4.4 and the documented installation and resolution checks pass.

由索引模型根据 Issue 内容生成。