Repository navigation
Reserved interface descriptions state their contracts directly (#582)
Reserved action, settings, and reusable-workflow descriptions state their existing contracts directly. Native help uses direct prose and the existing source-backed types. The two consent descriptions omit only the final appeal. No caller action is required.
Changed: Interface and consent descriptions use direct prose
The five reserved actions retain every condition, effect, default, expression, permission, output, exact example, version constraint, and required artifact-name match. WhatIf descriptions use the action only logs so the original log-only scope stays explicit. Native help retains its tags, associations, and all 16 exact examples.
The settings schema and reusable workflows change only existing description values. Each TOML edit removes only the final consent appeal; the preceding cookie/measurement statement and all consent settings remain exact. This PR does not certify that unverified statement or repair a runtime, schema, caller, output, recovery, or release contract.
Technical details
Current head: 896b048fcd7624abd00dddaf6d0f55424ab399d6 on interface-description-prose.
Reviewed source parent: 09d209ee07418174c6835cb64b6977c2740ad962.
Identical full tree: 8938314d16ecd805b595a0729b1f1c91ae77e0b2.
Fresh main base: 64ffb1d87f86bd5d487891fdb140b78385a00531.
Seven small prose commits and one source-identical empty trigger are pushed. The draft opened after the first coherent commit; the user later made it ready for review. The final source diff still contains 18 modified files, 176 insertions, and 151 deletions; it adds no repository file, test, enforcement tool, dependency, or audit Markdown.
The November 11 preparation first verifies clean local/origin/live source, the reviewed head and full tree, and zero exact-head native check runs, suites, statuses and PR-event runs after retention expiry. Only that observed missing condition permits the one ordinary empty commit. Its immediate parent, identical tree, zero-file delta and normal remote push are verified. No reset, rebase, amendment, force push, second trigger, or source edit occurs. The accepted source audit is reused for this equal tree, not repeated or represented as a new run.
Exact trigger proof, native PR-event CI, fresh review and bounded handoff evidence is saved on the completion Task and read back from GitHub.
The complete original source audit is attached to the closing Task:
- Full candidate inventory, no-change/protected classifications, command results, and precise factual follow-ups: 38 source files, all 163 description candidates, all 22 native-help units, and all 31 existing parameter descriptions.
- Exact description/consent changed regions: complete before/after text, immutable source citations, and preserved-meaning guards.
- Exact native-help changed regions: complete before/after text, immutable source citations, and preserved-meaning guards.
Together, the region tables record all 109 exact replacements. The description inventory contains 56 revised YAML/JSON values, 102 no-change values, three protected existing-contract values, and two final-appeal-only TOML edits.
| Changed surface | Standards checked | Framework/source evidence | Result |
|---|---|---|---|
| Five reserved action metadata files; settings schema; five reusable workflows | Pinned Natural Language, Writing Guide, YAML, GitHub Actions, Documentation; native JSON annotations | Existing action contracts, settings and workflow-input references; complete structured-node and exact raw-text comparisons | Prose aligned; every non-description node/value and exact constraint/example stays unchanged |
| Five PowerShell native-help source files | Pinned PowerShell Scripts/Functions, Documentation, Natural Language; native Get-Help format | Eight PowerShell candidates inspected; 22 native-help units and 16 examples | Exact program tokens/flags, AST shape, requirements, directives, BOM, help tags/association, syntax and parameter contracts preserved; zero parse/analyzer findings |
| Two existing consent descriptions | Pinned writing criteria; native TOML string format | Both complete TOML files parse; docs site builds from an exact session-local copy | Only the final appeal is removed; all 66 rendered pages retain the remaining statement and Accept/Reject labels |
Existing source validation: one-off exact raw-text/BOM/EOL/metadata comparisons pass for all 38 candidates; all 109 expected regions are present. Existing ConvertFrom-Yaml parses all ten changed YAML files. Existing Test-Json accepts the schema and anonymous empty settings. Invoke-ScriptAnalyzer reports zero findings on all eight PowerShell candidates before and after. zensical build --clean exits 0 with No issues found. git diff --check passes and the worktree is clean. These are the retained November 4 results for the identical source tree, not repeated November 11 audit runs.
Default YAML lint is not reported as clean: 152 baseline findings become 138 final findings, with no new rule/region finding. Existing formatting findings remain outside the released prose scope. No local publishing, cleanup, installation, network example, fixture, or benchmark script runs. No Markdown source or link changes; all source HTTP(S) links stay exact.
The fixed writing source is MSXOrg/docs@1b0d4893c8fefabe604d04393068a8ff6f848082, including the Documentation Model, Natural Language, Writing Guide, Documentation, Markdown, applicable native formats, and Task/PR formats. The official STE overview/FAQ and local Issue 9 rules/dictionary are consulted. Manual material stays local. No full STE-conformance claim is made.
Current main, #564's protected handback, and all first-pass path exclusions are checked before the prose commits. Every README, Markdown source, plugin, test, both functional fixture trees, first-pass #565-#570 path, and unmatched proposal-only file stays unchanged. Existing proposal contracts and branches are not amended or woken.
Retained facts: the cookie/measurement statement remains unverified; release-note fallback wording still needs owner reconciliation with the pushed-commit-message branch; three settings descriptions remain protected; the existing all-releases and retry-count terminology and imprecise array/PSSemVer help entries remain separate contract/help questions. The full audit gives exact sources and does not select a runtime or policy fix.
Fresh native review: one explicit request at 2026-11-11T14:57:49-14:57:51Z, timeline event 35811024275 at 14:57:52Z, produces review 5860771218, APPROVED at 15:00:10Z on exact head 896b048fcd7624abd00dddaf6d0f55424ab399d6. REST and GraphQL confirm its ID, timestamp, head and contributor-associated Copilot author. The old dismissed/rebound November 4 review 5769820091 is excluded. The fully paginated thread query after fresh submission has zero threads. No redundant same-head review round follows.
Native PR-event CI, bounded snapshot near 2026-11-11T15:00:40Z: Release, Test-Actions, Docs and Linter pass. The Default and WithManifest self-test runs remain in progress; the exact-head check list has 42 successes, four skips and six in-progress checks, with no completed failure. All six native runs associate pull_request, this exact head, PR582 and the recorded main base. Docs publication and index mutation are skipped on the PR event. No manual dispatch, unrelated run, fixture edit or workflow change is used. Pending jobs are not represented as passing.
Lifecycle: the PR remains open/non-draft. Agent merge remains enabled for review, CI and conflict maintenance only, with merge_pr=false; GitHub auto_merge=null. This preparation changes none of those settings, starts no watcher or schedule, and does not merge, publish, close the Task, or make a policy decision. Root alone controls the later guarded merge/publication sequence. The two incomplete native self-test runs remain the bounded handoff blocker.
The scoped convergence search covers the writing campaign and its source/contract siblings. Only the closing Task below is resolved by this delivery after eventual merge. Other campaign, caller, native-help, and output gaps remain non-closing context.