g-skl-platform-monitor

name: g-skl-platform-monitor description: Cross-platform health and freshness monitor for the registry-driven gald3r platform roster (PLATFORM_REGISTRY.yaml — single source of truth, T516). Checks per-platform capability gaps against the Cursor reference, scans official docs for breaking changes, validates platform-specific config, and generates the PLATFORM_STATUS / PLATFORM_CAPABILITY_MATRIX living indexes. Owned by g-agnt-platformer. token_budget: low subsystem_memberships: [PLATFORM_INTEGRATION]

Activate when checking platform capability gaps, scanning platform docs for breaking changes, validating platform config, or (re)generating PLATFORM_STATUS.md / PLATFORM_CAPABILITY_MATRIX.md. The g-agnt-platformer agent is the primary caller.

The Platform Roster — registry-driven (single source of truth, T516)

The roster of shipped platforms is defined once in gald3r_templates/gald3r_core/platforms/PLATFORM_REGISTRY.yaml. Every consumer derives from it — there is no hand-typed list anywhere else:

scripts/check_platform_status.py / .ps1 load KNOWN_PLATFORMS from the registry via the shared reader scripts/platform_registry.py (the .ps1 shells out to it so YAML parsing is not duplicated). If the registry file is missing, the reader returns a baked-in fallback roster so the tooling still runs.
The -GenerateMatrix path iterates the registry roster and resolves each platform's PLATFORM_SPEC.md from its registry spec_path, so PLATFORM_STATUS.md and PLATFORM_CAPABILITY_MATRIX.md share one roster.
A roster-parity gate (g-medic L1-J → g-skl-medic/scripts/check_roster_parity.py) asserts overlays == registry == specs == STATUS rows and fails loudly on drift.

Registry entry fields: name (canonical id = overlay dir), display_name, overlay_dir, spec_path, lifecycle (active|abandoned|off_target|redundant|stub), alias_of (e.g. vibe → mistral), support_level, notes. Edit the roster in the registry and nowhere else. To add a platform: add its overlay dir, add a registry entry, run the parity gate, then generate at least a stub PLATFORM_SPEC.md (honest all-❓). Aliases are resolved via alias_of and never double-counted.

Reference implementation = g-skl-platform-cursor. Source-of-truth per platform = g-skl-platform-<name>/PLATFORM_SPEC.md (and its docs_url: frontmatter). Status indexes live at .gald3r/PLATFORM_STATUS.md and .gald3r/PLATFORM_CAPABILITY_MATRIX.md.

Inspect the roster: python scripts/platform_registry.py --list (canonical names) / --list --all (incl. aliases) / --json (full registry).

Scaffolding note (T1460): the operation contracts below are fully specified. The freshness loop is now implemented (T513): SCAN_DOCS → spec proposals (spec_refresh.py, T514, GAP A) and GENERATE_STATUS (generate_status.py, T515, GAP B) close the broken crawl→spec→status chain. The remaining heavy CHECK gap-analysis / VALIDATE config-introspection logic is still deferred to the per-platform tasks (T1461–T1483); those spots are marked [deferred — T146x] in the body. This is scaffolding by design, not an unannotated stub.

Operations

CHECK `<platform>`

Compare one platform's declared capability support against the Cursor reference.

Read g-skl-platform-<platform>/SKILL.md — extract declared support for hooks, rules, skills, commands, MCP (sections 3/6/7/8 in the platform SKILL or its PLATFORM_SPEC).
Read g-skl-platform-cursor/SKILL.md as the reference baseline.
For each capability the reference has, check whether the platform declares an equivalent. Emit gaps: cursor has <X>; <platform> has no equivalent.
Update the platform's row in PLATFORM_STATUS.md (capability cells + Notes).
If g-skl-platform-<platform>/SKILL.md does not exist (antigravity) → report NO SKILL — create via T1465 and mark all cells ❓.

Delegates the read/report mechanics to the engine: gald3r platform status --platform <name> (falls back to scripts/check_platform_status.ps1 -Platform <name> where the engine is unavailable).

SCAN_DOCS `<platform>`

Crawl the platform's official docs and diff against the last crawl.

Read docs_url: from g-skl-platform-<platform>/SKILL.md frontmatter.
Invoke g-skl-crawl (or g-skl-recon-docs) on that URL; store results under {vault_location}/research/platforms/<platform>/.
Diff against the previous crawl snapshot. Surface changed sections: These sections changed since last scan — review for gald3r compatibility impact.
Feed the crawled doc snapshot to the spec-refresh consumer (T514, GAP A): scripts/spec_refresh.py --platform <name> --crawl-snapshot <export.json> [--crawl-ledger <registry.json>] (.ps1 parity wrapper alongside). It emits a reviewable proposal — a PLATFORM_SPEC.md.proposed draft + a "what changed and why" summary — and stamps the proposed last_doc_scan from the crawl-ledger completion date. It NEVER blind-overwrites the curated spec: capability-cell disagreements between the crawled docs and the spec surface as [needs-review] for a human to judge; only the mechanical last_doc_scan stamp lands on --apply. Then regenerate PLATFORM_STATUS.md via GENERATE_STATUS and the matrix via GENERATE_MATRIX.

SCAN_ALL

Run CHECK + SCAN_DOCS for every platform in the list, then summarize: N healthy (✅) · M need attention (⚠️) · K need full rework (❌) · U unknown (❓). Respects each platform's crawl_max_age_days — skips a doc scan that is still fresh.

VALIDATE `<platform>`

Confirm the platform's config is platform-specific, not Cursor-copied.

Inspect project_template/.gald3r_sys/platforms/.<platform>/ (the per-platform override dir).
Check format correctness: is hooks.json valid JSON? does settings.json match the platform's expected structure?
If the override dir is empty or its files are byte-identical to the Cursor copy → report Platform <X> config appears Cursor-generic — see gaps.
Record findings in the platform's PLATFORM_STATUS.md Notes. [deferred — T146x: per-platform structural schema checks]

GENERATE_MATRIX

(Re)build .gald3r/PLATFORM_CAPABILITY_MATRIX.md — the registry roster (PLATFORM_REGISTRY.yaml) × 6 capability columns (Hooks, Rules, Skills, Commands, MCP, Docs Fresh). Cells: ✅ / ⚠️ / ❌ / ❓. Source the cell values from each platform's PLATFORM_SPEC.md (resolved via the registry spec_path). Generated, never hand-maintained.

GENERATE_STATUS

(Re)build .gald3r/PLATFORM_STATUS.md from the specs + the crawl ledger (T515, GAP B): scripts/generate_status.py --apply [--crawl-ledger <registry.json>] (.ps1 parity wrapper alongside; dry-run is the default — omit --apply to preview). Closes the second freshness-loop gap: STATUS was hand-maintained and rotted (check_platform_status reads it READ-ONLY and never wrote it). Source-of-truth = Option 2 merge: the curated Status verdict + Notes columns are PRESERVED from the existing STATUS; only the mechanical cells are regenerated — the 5 capability cells derived from each PLATFORM_SPEC.md ## Capability Summary the SAME way GENERATE_MATRIX derives them (so a regen leaves zero STATUS-vs-matrix cross-check warnings), and Last Doc Scan taken from the crawl ledger (real update_crawl_registry completion date) else the spec frontmatter last_doc_scan (never "now" blindly). Idempotent: a re-run with no input change is byte-identical modulo the generated timestamp line (use --no-timestamp for byte-for-byte CI diffs). The human edits the SPEC, not STATUS.

UPGRADE `<platform>`

Given a SCAN_DOCS result, propose specific config changes to gald3r's platform template.

Read the doc-scan diff for the platform.
Map changed capabilities to gald3r template locations (common skill vs. platforms/.<platform>/ override) using the decision tree in g-skl-platform-cursor.
Produce a proposal/diff for human review — never auto-apply, never run platform_parity_sync.ps1. The human reviews, then runs parity sync. [deferred — T146x: per-platform change mapping]

Status Legend

Symbol	Meaning
✅	Verified working (with evidence)
⚠️	Partial / Cursor-generic / untested-but-present
❌	Not supported on this platform
❓	Unknown — never checked

Wiring

Agent owner: g-agnt-platformer.
Commands: @g-platform-check, @g-platform-scan-docs, @g-platform-status.
Engine: gald3r platform status [--platform <name>] (CHECK entry point); fallback scripts/check_platform_status.ps1.
Freshness loop (T513): scripts/spec_refresh.py/.ps1 (T514 — crawled docs → PLATFORM_SPEC.md proposals) and scripts/generate_status.py/.ps1 (T515 — specs + crawl ledger → PLATFORM_STATUS.md). Shared spec/ledger parsing in scripts/platform_spec_io.py. Both run host-side (C-001), need no DB connection or migration, and are dry-run by default (proposals, not blind writes).
Roster source of truth: gald3r_templates/gald3r_core/platforms/PLATFORM_REGISTRY.yaml (T516), read via scripts/platform_registry.py.
Roster-parity gate: g-skl-medic/scripts/check_roster_parity.py, wired into g-medic L1-J — fails loudly when overlays / registry / specs / STATUS rows disagree.
Medic: g-medic L2 calls g-skl-platform-monitor CHECK <current-platform> for platform health.

g-skl-platform-monitor