> ## Documentation Index
> Fetch the complete documentation index at: https://docs.agentlinkops.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Local context result conventions

> Read repository context results, keep manual judgments separate from saved facts, and handle partial refreshes, missing files and unresolved citations.

Local context tools return their JSON result in `structuredContent` and in a JSON text content block. These records describe files inside the host's repository root. They do not use the hosted command response envelope.

A missing manual or profile file returns an `error` field and sets MCP `isError`. An unresolved citation also sets `isError` while preserving the citation report. A thrown service or path-boundary error returns text with `isError`; do not assume every failed call has `structuredContent`.

## Results by tool

| Tool                                                                | What to read                                                                 |
| ------------------------------------------------------------------- | ---------------------------------------------------------------------------- |
| [context\_status](/reference/context/context_status)                | Project context, sanitized grant state, manual presence and local row counts |
| [context\_manual](/reference/context/context_manual)                | Manual statements and nominated pages with no invented observations          |
| [context\_focus](/reference/context/context_focus)                  | Manual-first page selection with saved observations kept in separate fields  |
| [context\_gsc\_page](/reference/context/context_gsc_page)           | Page/query metrics, date window, coverage state and nullable totals          |
| [context\_gsc\_refresh](/reference/context/context_gsc_refresh)     | Run, property and window outcomes; request budgets; rows appended            |
| [context\_gsc\_import](/reference/context/context_gsc_import)       | Accepted rows, every refused row and preserved connector identities          |
| [context\_profile\_build](/reference/context/context_profile_build) | Bounded robots, sitemap and selected-page outcomes with explicit limits      |
| [context\_profile\_check](/reference/context/context_profile_check) | Citation count, unresolved references and read-only fact status              |

Each tool page includes its nested result fields and a fixture-backed example. Examples demonstrate shape and semantics; their dates and counts are not account evidence. Profile examples use fixture fetches rather than a claim of live web coverage.

## Preserve uncertainty and partial work

`manual_only` is a valid configuration. A missing Google grant does not prevent your agent from using manual site inputs. Null observation, inference and recommendation fields preserve their separate ownership.

Read per-property and per-window refresh outcomes even when the run state is `ok`. A later quota refusal or authorization loss leaves earlier appended rows intact. A profile build can complete while individual pages remain unreadable. Keep those page outcomes and cap counters in the next report.

A missing page in bounded search rows never means zero impressions. Click-through rate is derived from saved counts; average position remains an average. Keep unknown metrics null and preserve source dates. Project configuration and saved handoff state can contain customer-owned extensions; connection status returns only selected sanitized fields.

Use [repository context setup](/guides/local-context) for the fixed root boundary, [common schemas](/reference/schemas) for shared conventions and [troubleshooting](/guides/troubleshooting) for path and grant recovery.