docs(agents): compact root guidance and preserve task-specific knowledge (#1645)

* docs(agents): compact root guidance and preserve task-specific knowledge

* fix(agents): parse guidance navigation with Markdown tokens

* fix(agents): validate generic literal repository paths

* fix(agents): distinguish code symbols and shortcut images

* fix(agents): recognize SCSS filename literals

* fix(agents): handle fenced imports and encoded paths

* fix(agents): parse prose and rendered HTML anchors

* fix(agents): validate rendered HTML navigation

* fix(agents): use GitHub-compatible heading slugs

* fix(agents): require standalone top-level Claude import

* fix(agents): exclude HTML-contained guidance imports

* fix(agents): handle image fragments and quoted imports

* fix(agents): validate visible HTML and image source sets

* fix(agents): recognize package scopes and route source work

* fix(agents): parse JSONC and constrain package exemptions

* fix(agents): decode link entities and allow package subpaths

* fix(agents): route source work and check extensionless files

* fix(agents): support package versions and source fragments

* fix(agents): accept qualified package prose

* fix(agents): retain rendered context for Markdown references

* fix(agents): validate visible headings and spaced paths

* fix(agents): validate media and hyphenated literal paths

* fix(agents): decode full HTML entities and media assets

* fix(agents): recognize possessive package mentions

* fix(agents): validate extensionless imports and version comparators

* fix(agents): retain visible backticks and explicit path punctuation

* fix(agents): validate image-map navigation targets

* fix(agents): count all Markdown line endings in budgets

* fix(agents): delimit package prose at Unicode punctuation

* fix(agents): normalize punctuation for extensionless imports

* fix(agents): preserve filenames across prose punctuation

* fix(agents): validate iframe document references

* fix(agents): inspect document suffix before URL fragments

* fix(agents): unify Markdown suffix and encoded import guards

* fix(agents): handle wildcard versions and alternate documents

* fix(agents): validate document formats and trim HTML URLs

* fix(agents): cover document families and guidance basenames

* fix(agents): require files for media references

* fix(agents): preserve block boundaries and validate embeds

* fix(agents): normalize internal HTML URL whitespace

* fix(agents): reject empty media and ignore URL at-signs

* fix(agents): validate srcdoc references and empty srcset

* fix(agents): honor HTML bases and preserve adjacent imports

* fix(agents): convert base file URLs to native paths

* fix(agents): preserve imports after bare URL punctuation

* fix(agents): exclude opaque URI prose from import scans

* fix(agents): keep import tokens outside URI scheme matches

* fix(agents): restrict opaque URI exemptions to parsed links

* fix(agents): handle opening prose delimiters

* fix(agents): scan nested imports and share document suffixes

* fix(agents): reject pathless media and direct file URLs

* fix(agents): reject file bases and preserve quoted URL boundaries

* fix(agents): distinguish URL quotes and cover guidance variants

* fix(agents): validate SVG images and conventional guides

* fix(agents): handle declared package names handles and SVG use

* fix(agents): normalize closing punctuation on federated handles

* fix(agents): normalize Unicode punctuation on handles

* fix(agents): normalize possessive federated handles

* fix(agents): separate parenthetical prose from handles

* fix(agents): exclude www autolinks from import scanning

* ci: allow manual CodeQL validation of PR branches

* fix(agents): reject nonportable Windows drive links

docs/architecture/workspace-dashboard.md

@@ -327,3 +327,13 @@ Intentionally out of scope:
 2. Freeform widget grid with collision management.
 3. External data rails such as RSS, sports, or news adapters.
 4. Per-user A/B variants of rail ordering.
+
+## Source subscription expiry
+
+Source cards show a passive subscription-expiry chip: amber within seven days,
+error-toned once expired. Account details stay behind the Account info menu.
+`DashboardSourceExpiryService` in `libs/workspace/dashboard/data-access` reads
+Xtream expiry from cached `PortalStatusService.checkPortalStatusDetails()`
+(`exp_date`). Stalker uses the persisted `stalkerAccountInfo` snapshot from the
+playlist payload, not the metadata row; each source therefore needs one memoized
+full-playlist read. The chip is not a separate account-refresh request.