Skip to content

[finding] SKILL_MAP census residue: objectstack-api advertises api/rest-server.zod.ts its SKILL.md never mentions, and objectstack-platform's index omits stack.zod.ts, the shape its body spends the most words on #15264

Description

@os-justin

Residue of the census run for #14462. That card carries the census across all nine SKILL_MAP entries and applies one uniform ruling; the family PR acted on the four rows its members adjudicated (objectstack-ai, objectstack-automation, objectstack-formula, objectstack-i18n). These two rows came out of the same sweep with the same shape and no card has adjudicated them, so the PR recorded them and edited nothing. Filed unassigned.

Measured at origin/main 9c1bcda3

Direction 1 — advertised, never taught

package pointer evidence
objectstack-api api/rest-server.zod.ts zero mentions in skills/objectstack-api/SKILL.md of the file, of RestApiConfig / CrudEndpointsConfig / MetadataEndpointsConfig / BatchEndpointsConfig / RouteGenerationConfig / RestServerConfig, or of the words "rest server". Swept over the skill's whole authored markdown, not the SKILL.md alone.

Every other core entry in that list is reachable from the body by at least the schema's own exported names.

Direction 2 — taught, never advertised

package schema evidence
objectstack-platform stack.zod.ts the SKILL.md's subject is defineStack, and it names ObjectStackDefinitionSchema, ObjectStackDefinition, ObjectStackDefinitionInput and ObjectStack in its own prose. The file is in no package's core list and in no index's closure.

Same shape as the ai/solution-blueprint.zod.ts half of #14462: the body names the schema, the catalog has no pointer to it.

One row deliberately NOT filed here

kernel/plugin-capability.zod.ts in the objectstack-platform list scored as advertised-not-taught on a mechanical sweep, and a hand read overturns it: the SKILL.md teaches requires: and capability resolution at length. It routes readers to kernel/platform-capabilities.ts for the token list rather than to this file's PluginCapabilityManifestSchema, so the pointer is thin rather than wrong. Recorded so the next census does not re-raise it.

Why this is a separate card

The uniform ruling on #14462 is "narrow the map to what the body teaches", and applying it here is defensible — but the four adjudicated rows each came with a measured consumer census or an explicit ADR retirement, and these two have only the sweep. Deleting a published pointer, or adding one, on a governed customer surface is worth its own decision rather than a rider on a PR whose reviewer is checking four other things.

Refs: #14462 · #14466 · #14562 · #14565

Dedupe: one targeted search_issues over this repo (repo-scoped REST is 403 for this seat), validated in-session by a control query that returned its four known hits. Nothing open covers either row.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions