Skip to content

Latest commit

 

History

History
86 lines (67 loc) · 3.69 KB

README.md

File metadata and controls

86 lines (67 loc) · 3.69 KB

Autonomous Run Governance Preset

Version: 0.1.4 Requires: spec-kit >= 0.8.3 Recommended priority: 70

This optional preset governs complete, delegated Spec Kit runs without making remote authority implicit. It composes with security, architecture, iSAQB, accessibility, cross-platform, and agent-parity presets.

It provides:

  • speckit.autonomous for end-to-end, convergence-based feature delivery
  • speckit.autonomous-retrospective for classified workflow learning
  • Constitution, Spec, Plan, Tasks, and Agent addenda
  • runbook, evidence, retrospective, and readiness-checklist templates
  • machine-readable gate-requirements and exact-head evidence templates
  • read-only Bash and PowerShell gate-evidence validators
  • field-validation evidence from six product runs plus audit and closure runs

LocalImplementation is the safe default. PublishPR and MergeAndSync require explicit current authorization. Provider bypass is never inferred from autonomy, repository access, or an earlier run.

The preset does not prescribe a programming language, test framework, CI provider, branch version scheme, agent vendor, or merge policy. Project rules remain authoritative and supply the concrete commands and evidence paths.

Development install:

specify preset add --dev /path/to/autonomous-run-governance --priority 70

Published install:

specify preset add --from https://github.com/hindermath/spec-kit-preset-autonomous-run-governance/archive/refs/tags/v0.1.4.zip --priority 70

Version 0.1.4 retains the machine-checkable gate contract introduced in v0.1.3 and makes its invocation stable after ZIP installation. Declare every acceptance gate before implementation with autonomous-run-gate-requirements-template, then collect exact-head execution evidence with autonomous-run-gate-evidence-template. The requirements define command and optional runner/platform tokens, so a green tooling-only job cannot satisfy a runtime gate merely because its name contains a platform.

Validate the final reviewed head on macOS/Linux:

bash .specify/presets/autonomous-run-governance/scripts/validate-autonomous-gate-evidence.sh \
  --requirements specs/NNN-feature/autonomous-gate-requirements.json \
  --evidence /temporary/provider-gate-evidence.json \
  --head "$(git rev-parse HEAD)"

The Bash validator requires python3 or python for strict JSON parsing. On Windows use PowerShell 7:

pwsh -NoProfile -File .specify/presets/autonomous-run-governance/scripts/validate-autonomous-gate-evidence.ps1 `
  -Requirements specs/NNN-feature/autonomous-gate-requirements.json `
  -Evidence $env:TEMP/provider-gate-evidence.json `
  -Head (git rev-parse HEAD)

Dot-source the PowerShell script to use the equivalent Advanced Function Test-AutonomousGateEvidence in an existing session.

Always invoke the installed scripts through bash or pwsh -File. Git records the Bash source as executable, but a preset installer or ZIP extraction may copy it without that mode bit. Direct path execution is therefore not a portable contract.

The accepted requirements file is a reviewed feature artifact. The provider evidence file SHOULD be generated in a temporary directory after checks finish, because committing exact-head evidence before merge would create a new head and self-invalidate the proof. A causal closeout may record the immutable run and validator result after delivery without claiming that the closeout proves its own new revision.

Validate with specify preset list, specify preset info autonomous-run-governance, and specify preset resolve for the affected templates. Installation does not itself execute a feature or grant remote write, merge, bypass, or provider-administration rights.