Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
30 changes: 23 additions & 7 deletions .agents/skills/speckit-autonomous/SKILL.md
Original file line number Diff line number Diff line change
Expand Up @@ -18,9 +18,10 @@ stages; this skill only orchestrates them and enforces cross-stage gates.
3. Determine the delegated delivery mode from the user's request:
`LocalImplementation`, `PublishPR`, or `MergeAndSync`. Never infer remote
write or merge authority from general autonomy alone.
4. Confirm intake ordering, branch identity, clean ownership boundaries, and
the six installed governance presets. Stop only for a material conflict,
missing required authority, destructive ambiguity, or an unmet hard gate.
4. Confirm intake ordering, branch identity, clean ownership boundaries, the
six baseline governance presets, and optional `autonomous-run-governance`
v0.1.4. Stop only for a material conflict, missing required authority,
destructive ambiguity, or an unmet hard gate.
5. Create the numbered feature branch through `speckit-git-feature` when a new
feature is required. Resume the existing branch and feature directory when
the run already exists.
Expand All @@ -45,6 +46,12 @@ Create `specs/NNN-feature/pr-evidence.md` from
implementation edit. Keep scope, decisions, commands, results, skipped
triggers, residual risks, review state, and follow-ups current during the run.

Before implementation, resolve
`autonomous-run-gate-requirements-template` and create the reviewed
`specs/NNN-feature/autonomous-gate-requirements.json`. Give every acceptance
gate a stable ID, required scope, command tokens, optional runner/platform
tokens, and either `Applicable` or a justified `N/A` with re-evaluation trigger.

## Converge

- Clarification converges when no remaining question would materially change
Expand All @@ -59,10 +66,12 @@ triggers, residual risks, review state, and follow-ups current during the run.
- Remote review converges when required checks pass and no actionable review
thread remains. Record unavailable reviewers or quota failures as evidence;
never represent them as successful reviews.
- Before merge, map every acceptance-specific gate to the actual workflow, job,
platform, and command that produced its evidence. A green aggregate or an OS
name is insufficient when the job did not execute the required runtime,
platform, documentation, security, or other proof. Missing scope blocks merge.
- Before merge, derive every acceptance-specific gate's command and runner from
workflow definitions or job logs, never from a green aggregate, workflow,
job, or OS-shaped name. Create temporary provider-neutral exact-head evidence
from `autonomous-run-gate-evidence-template` and run the installed Bash or
PowerShell validator. Missing, stale, contradictory, empty, duplicate-Primary,
or token-mismatched evidence blocks merge; bypass supplies no proof.

## Shape Work

Expand Down Expand Up @@ -116,6 +125,13 @@ channel contains no PowerShell error record, shell command-not-found message, or
equivalent fatal signature. A nominal zero exit status does not override such an
error; record the helper as failed and route the defect to bounded remediation.

For `PublishPR` or `MergeAndSync`, validate the final reviewed head with
`bash .specify/presets/autonomous-run-governance/scripts/validate-autonomous-gate-evidence.sh`
on macOS/Linux or `pwsh -NoProfile -File` with the matching `.ps1` on Windows;
installed mode bits are not a portable contract. The validator is read-only and
grants no remote authority. Keep the provider evidence snapshot temporary before
merge; committing it creates a new head and self-invalidates its reviewed-head claim.

## Deliver

Use logical Git checkpoints only after their acceptance gates pass. On numbered
Expand Down
20 changes: 20 additions & 0 deletions .claude/skills/speckit-autonomous/SKILL.md
Original file line number Diff line number Diff line change
Expand Up @@ -66,6 +66,16 @@ authority from general autonomy.
restore the original index state.
10. Route out-of-scope findings to named follow-up evidence instead of silently
expanding the feature.
11. Before implementation, resolve the preset's gate-requirements template and
declare every acceptance gate in a reviewed feature artifact. Applicable
gates name required command tokens and any runner or platform tokens; `N/A`
gates name their rationale and re-evaluation trigger.
12. Before merge, generate provider-neutral gate evidence for the exact current
reviewed head in a temporary location and run the installed validator through
`bash <validator.sh>` or `pwsh -NoProfile -File <validator.ps1>`; installers
may not preserve executable mode bits. Every gate needs exactly one Primary row;
supplemental rows must point to that primary evidence. Missing, stale,
contradictory, empty, or token-mismatched evidence blocks merge.

## Remote Closeout

Expand All @@ -79,6 +89,16 @@ or platform, and command that actually executed it. A green aggregate or a
platform-named job proves only its executed scope. Missing required proof
blocks merge and cannot be replaced by a permission or ruleset bypass.

Derive `executedCommand` and `runnerOrPlatform` from the workflow definition or
job log, not from a check, workflow, or job name. The read-only validator checks
the accepted requirements hash, full reviewed head, gate completeness,
Applicable/N/A boundaries, required command and runner tokens, and Primary versus
Supplemental evidence. Its successful exit grants no remote authority.

Keep exact-head provider evidence temporary during the merge decision. Committing
that evidence would create a new head and invalidate its own reviewed-head claim.
Use the existing causal-closeout boundary for post-delivery facts.

Use one pre-named causal closeout only when current-head or post-merge facts
cannot be committed without invalidating themselves. Keep it evidence-only and
single-commit-capable; verify its own terminal provider facts externally.
Expand Down
4 changes: 2 additions & 2 deletions .github/agents/copilot-instructions.md
Original file line number Diff line number Diff line change
Expand Up @@ -278,12 +278,12 @@ GitHub Pages is published from `.github/workflows/pages.yml`: build root `docfx.
- Repository-Prüfhelfer erhalten den Repository-Root explizit. Ein Pass benötigt erwarteten Exitcode und einen Fehlerkanal ohne PowerShell-ErrorRecord, `command not found` oder gleichwertige fatale Signatur.
- Vor einem Commit wird der beabsichtigte Lieferkandidat separat gestaged und mit `git diff --cached --check` sowie einem Pfadvergleich gegen den Repository-Status geprüft; `git diff --check` allein erfasst keine ungetrackten Dateien, und fremde Änderungen bleiben ausgeschlossen.
- Vor einem Merge wird jedes Acceptance-Gate dem tatsächlichen Workflow, Job, Betriebssystem und ausgeführten Command zugeordnet; ein grüner Aggregatstatus oder ein Plattformname gilt nicht als Proof, wenn der Job den geforderten Runtime-, Plattform-, Dokumentations- oder Security-Nachweis nicht ausgeführt hat.
- Optional installiert: `autonomous-run-governance` v0.1.2 prio 70 aus dem öffentlichen Tag-ZIP. Der projektgebundene Codex-Skill `$speckit-autonomous` bleibt am selben eindeutigen Pfad als lokaler Override bestehen, weil er TuiVision-spezifische Nummerierungs-, Build-Zähler-, DocFX-/A11Y- und historische Source-Verträge ergänzt; Preset-Command, Retrospektiv-Skill, Runbook und Adoption-Evidence bleiben die portablen beziehungsweise gemeinsamen Nachweisflächen. Version 0.1.2 prüft zusätzlich den exakt beabsichtigten Staging-Kandidaten und ordnet jedes Acceptance-Gate dem wirklich ausgeführten Workflow, Job, Runner beziehungsweise der Plattform und dem Befehl zu; grüne Namen und Bypass ersetzen keinen technischen Nachweis.
- Optional installiert: `autonomous-run-governance` v0.1.4 prio 70 aus dem öffentlichen Tag-ZIP. Der projektgebundene Codex-Skill `$speckit-autonomous` bleibt am selben eindeutigen Pfad als lokaler Override bestehen, weil er TuiVision-spezifische Nummerierungs-, Build-Zähler-, DocFX-/A11Y- und historische Source-Verträge ergänzt; Preset-Command, Retrospektiv-Skill, Runbook und Adoption-Evidence bleiben die portablen beziehungsweise gemeinsamen Nachweisflächen. Version 0.1.4 verlangt vor der Implementierung deklarierte Acceptance-Gates und prüft vor dem Merge temporäre Evidence für den exakten HEAD. Der Validator gleicht Requirements-Hash, HEAD, Command-/Runner-Tokens und genau einen `Primary`-Nachweis mit echten Workflow-Definitionen oder Job-Logs ab; grüne Namen, Validator und Bypass ersetzen weder technischen Nachweis noch Remote- oder Merge-Berechtigung.
- Jeder Lauf schützt den akzeptierten Scope, verwendet triggerbasierte Validierung und dokumentiert eine kurze Retrospektive für spätere Runbook-Verfeinerungen.

*Fully delegated Spec-Kit runs follow `docs/spec-kit-autonomous-runbook.md` and use the repository-local `$speckit-autonomous` skill. Determine `LocalImplementation`, `PublishPR`, or `MergeAndSync` from the current user request; general autonomy does not silently grant remote write or merge authority. Create evidence before implementation, iterate optional stages to their defined convergence criteria, prove one representative vertical slice before broad rollout, serialize shared writers, require each remote or delivery task to name its exact repository evidence path, route self-invalidating reviewed-head facts to one named closeout path after verifying them before merge, validate the exact staged candidate with `git diff --cached --check` so new files are included without unrelated changes, map each acceptance gate to the actual workflow, job, platform, and executed command before merge, protect accepted scope, use trigger-based validation, and record a short retrospective for later workflow refinement.*

*Optional installed preset: `autonomous-run-governance` v0.1.2 at priority 70 comes from the public tag ZIP. The project-owned Codex `$speckit-autonomous` skill stays at its single path as a local override because it adds TuiVision-specific numbering, build-counter, DocFX/A11Y, and historical-source contracts. The preset command, retrospective skill, runbook, and adoption evidence remain the portable or shared proof surfaces. Version 0.1.2 additionally validates the exact intended staged candidate and maps every acceptance gate to the workflow, job, runner or platform, and command that actually executed it; green names and bypass do not replace technical proof.*
*Optional installed preset: `autonomous-run-governance` v0.1.4 at priority 70 comes from the public tag ZIP. The project-owned Codex `$speckit-autonomous` skill stays at its single path as a local override because it adds TuiVision-specific numbering, build-counter, DocFX/A11Y, and historical-source contracts. The preset command, retrospective skill, runbook, and adoption evidence remain the portable or shared proof surfaces. Version 0.1.4 requires acceptance gates to be declared before implementation and validates temporary exact-HEAD evidence before merge. The validator checks the requirements hash, HEAD, command and runner tokens, and exactly one `Primary` proof against actual workflow definitions or job logs; green names, the validator, and bypass grant neither technical proof nor remote or merge authority.*
- Aktive Google-Agentenoberfläche ist Antigravity CLI mit Befehl `agy` und Spec-Kit-Integration `agy`. `GEMINI.md` und `~/.gemini/antigravity-cli/` bleiben Antigravity-kompatible Oberflächen; direkte `gemini`-Befehle sind nur historische oder ausdrücklich benötigte Enterprise-/API-Kompatibilität und keine lokale Pflicht.

*The active Google agent surface is Antigravity CLI through the `agy` command and Spec Kit `agy` integration. `GEMINI.md` and `~/.gemini/antigravity-cli/` remain Antigravity-compatible surfaces; direct `gemini` commands are historical or explicitly required enterprise/API compatibility, not a local requirement.*
Expand Down
20 changes: 20 additions & 0 deletions .github/agents/speckit.autonomous.agent.md
Original file line number Diff line number Diff line change
Expand Up @@ -63,6 +63,16 @@ authority from general autonomy.
restore the original index state.
10. Route out-of-scope findings to named follow-up evidence instead of silently
expanding the feature.
11. Before implementation, resolve the preset's gate-requirements template and
declare every acceptance gate in a reviewed feature artifact. Applicable
gates name required command tokens and any runner or platform tokens; `N/A`
gates name their rationale and re-evaluation trigger.
12. Before merge, generate provider-neutral gate evidence for the exact current
reviewed head in a temporary location and run the installed validator through
`bash <validator.sh>` or `pwsh -NoProfile -File <validator.ps1>`; installers
may not preserve executable mode bits. Every gate needs exactly one Primary row;
supplemental rows must point to that primary evidence. Missing, stale,
contradictory, empty, or token-mismatched evidence blocks merge.

## Remote Closeout

Expand All @@ -76,6 +86,16 @@ or platform, and command that actually executed it. A green aggregate or a
platform-named job proves only its executed scope. Missing required proof
blocks merge and cannot be replaced by a permission or ruleset bypass.

Derive `executedCommand` and `runnerOrPlatform` from the workflow definition or
job log, not from a check, workflow, or job name. The read-only validator checks
the accepted requirements hash, full reviewed head, gate completeness,
Applicable/N/A boundaries, required command and runner tokens, and Primary versus
Supplemental evidence. Its successful exit grants no remote authority.

Keep exact-head provider evidence temporary during the merge decision. Committing
that evidence would create a new head and invalidate its own reviewed-head claim.
Use the existing causal-closeout boundary for post-delivery facts.

Use one pre-named causal closeout only when current-head or post-merge facts
cannot be committed without invalidating themselves. Keep it evidence-only and
single-commit-capable; verify its own terminal provider facts externally.
Expand Down
4 changes: 2 additions & 2 deletions .github/copilot-instructions.md
Original file line number Diff line number Diff line change
Expand Up @@ -537,9 +537,9 @@ For GitLab repositories, use the authenticated `glab` CLI first for equivalent a
## Spec-Kit-Preset-Pflege / Spec Kit Preset Maintenance

- Standard-Preset-Set: `security-governance` v0.6.0 prio 10, `architecture-governance` v0.5.0 prio 20, `isaqb-architecture-governance` v0.2.0 prio 30, `a11y-governance` v0.4.0 prio 40, `cross-platform-governance` v0.2.0 prio 50, `agent-parity-governance` v0.3.0 prio 60.
- Optional installiert: `autonomous-run-governance` v0.1.2 prio 70 aus dem öffentlichen Tag-ZIP. Der projektgebundene Codex-Skill `$speckit-autonomous` bleibt am selben eindeutigen Pfad als lokaler Override bestehen, weil er TuiVision-spezifische Nummerierungs-, Build-Zähler-, DocFX-/A11Y- und historische Source-Verträge ergänzt; Preset-Command, Retrospektiv-Skill, Runbook und Adoption-Evidence bleiben die portablen beziehungsweise gemeinsamen Nachweisflächen. Version 0.1.2 prüft zusätzlich den exakt beabsichtigten Staging-Kandidaten und ordnet jedes Acceptance-Gate dem wirklich ausgeführten Workflow, Job, Runner beziehungsweise der Plattform und dem Befehl zu; grüne Namen und Bypass ersetzen keinen technischen Nachweis.
- Optional installiert: `autonomous-run-governance` v0.1.4 prio 70 aus dem öffentlichen Tag-ZIP. Der projektgebundene Codex-Skill `$speckit-autonomous` bleibt am selben eindeutigen Pfad als lokaler Override bestehen, weil er TuiVision-spezifische Nummerierungs-, Build-Zähler-, DocFX-/A11Y- und historische Source-Verträge ergänzt; Preset-Command, Retrospektiv-Skill, Runbook und Adoption-Evidence bleiben die portablen beziehungsweise gemeinsamen Nachweisflächen. Version 0.1.4 verlangt vor der Implementierung deklarierte Acceptance-Gates und prüft vor dem Merge temporäre Evidence für den exakten HEAD. Der Validator gleicht Requirements-Hash, HEAD, Command-/Runner-Tokens und genau einen `Primary`-Nachweis mit echten Workflow-Definitionen oder Job-Logs ab; grüne Namen, Validator und Bypass ersetzen weder technischen Nachweis noch Remote- oder Merge-Berechtigung.

*Optional installed preset: `autonomous-run-governance` v0.1.2 at priority 70 comes from the public tag ZIP. The project-owned Codex `$speckit-autonomous` skill stays at its single path as a local override because it adds TuiVision-specific numbering, build-counter, DocFX/A11Y, and historical-source contracts. The preset command, retrospective skill, runbook, and adoption evidence remain the portable or shared proof surfaces. Version 0.1.2 additionally validates the exact intended staged candidate and maps every acceptance gate to the workflow, job, runner or platform, and command that actually executed it; green names and bypass do not replace technical proof.*
*Optional installed preset: `autonomous-run-governance` v0.1.4 at priority 70 comes from the public tag ZIP. The project-owned Codex `$speckit-autonomous` skill stays at its single path as a local override because it adds TuiVision-specific numbering, build-counter, DocFX/A11Y, and historical-source contracts. The preset command, retrospective skill, runbook, and adoption evidence remain the portable or shared proof surfaces. Version 0.1.4 requires acceptance gates to be declared before implementation and validates temporary exact-HEAD evidence before merge. The validator checks the requirements hash, HEAD, command and runner tokens, and exactly one `Primary` proof against actual workflow definitions or job logs; green names, the validator, and bypass grant neither technical proof nor remote or merge authority.*
- Aktive Google-Agentenoberfläche ist Antigravity CLI mit Befehl `agy` und Spec-Kit-Integration `agy`. `GEMINI.md` und `~/.gemini/antigravity-cli/` bleiben Antigravity-kompatible Oberflächen; direkte `gemini`-Befehle sind nur historische oder ausdrücklich benötigte Enterprise-/API-Kompatibilität und keine lokale Pflicht.

*The active Google agent surface is Antigravity CLI through the `agy` command and Spec Kit `agy` integration. `GEMINI.md` and `~/.gemini/antigravity-cli/` remain Antigravity-compatible surfaces; direct `gemini` commands are historical or explicitly required enterprise/API compatibility, not a local requirement.*
Expand Down
20 changes: 20 additions & 0 deletions .opencode/command/speckit.autonomous.md
Original file line number Diff line number Diff line change
Expand Up @@ -63,6 +63,16 @@ authority from general autonomy.
restore the original index state.
10. Route out-of-scope findings to named follow-up evidence instead of silently
expanding the feature.
11. Before implementation, resolve the preset's gate-requirements template and
declare every acceptance gate in a reviewed feature artifact. Applicable
gates name required command tokens and any runner or platform tokens; `N/A`
gates name their rationale and re-evaluation trigger.
12. Before merge, generate provider-neutral gate evidence for the exact current
reviewed head in a temporary location and run the installed validator through
`bash <validator.sh>` or `pwsh -NoProfile -File <validator.ps1>`; installers
may not preserve executable mode bits. Every gate needs exactly one Primary row;
supplemental rows must point to that primary evidence. Missing, stale,
contradictory, empty, or token-mismatched evidence blocks merge.

## Remote Closeout

Expand All @@ -76,6 +86,16 @@ or platform, and command that actually executed it. A green aggregate or a
platform-named job proves only its executed scope. Missing required proof
blocks merge and cannot be replaced by a permission or ruleset bypass.

Derive `executedCommand` and `runnerOrPlatform` from the workflow definition or
job log, not from a check, workflow, or job name. The read-only validator checks
the accepted requirements hash, full reviewed head, gate completeness,
Applicable/N/A boundaries, required command and runner tokens, and Primary versus
Supplemental evidence. Its successful exit grants no remote authority.

Keep exact-head provider evidence temporary during the merge decision. Committing
that evidence would create a new head and invalidate its own reviewed-head claim.
Use the existing causal-closeout boundary for post-delivery facts.

Use one pre-named causal closeout only when current-head or post-merge facts
cannot be committed without invalidating themselves. Keep it evidence-only and
single-commit-capable; verify its own terminal provider facts externally.
Expand Down
Loading
Loading