console

name	console
description	Terminal-first (with an --html mode) report that composes agentm's existing observability surfaces — the health index + family table, queue-status-lite, the board-drift one-liner, spend from the observability rollup, and a memory-activity section (inbox, watchlist, incubator, newest curated entries, heat-policy decay report) — behind one entry point. Builds nothing new underneath; every section reads or shells out to a surface that already exists and is independently invocable on its own.
kind	skill
supported_hosts	["claude-code","antigravity"]
version	0.1.0
install_scope	project

console — the unified observability report

/console answers "what is agentm doing right now?" in one read — instead of running scripts/status.py, scripts/queue_status_lite.py, a board-drift check, an observability-console render, and a memory-vault glance separately. It is a static, on-demand report, not a live dashboard service — the same "static derived data needs no daemon" call the Autonomy arc's design already made for the observability console this skill extends (wiki/designs/agentm-autonomy.md, "Out of scope").

Not for: a live/continuously-refreshing view (there is no daemon behind this — re-run the skill to refresh), or a substitute for any of the deeper commands it composes (/doctor for install-health, /memory for the memory engine's own sub-commands, /report-board-drift for the operator-confirmed drift-correction cycle). /console is the glance; the composed commands are where you go deeper.

Modes

Mode	What it does
python3 harness/skills/console/scripts/console.py (or the installed equivalent, e.g. .claude/skills/console/scripts/console.py)	Terminal report: five sections (Health, Plans, Board drift, Spend, Memory activity), each on its own heading.
python3 .../console.py --html [--output PATH]	Renders the same five sections as a single self-contained static HTML page, extending scripts/health/observability_console.py's already-built spend page (its exact rendered tables are reused, not re-implemented) with the other four sections wrapped around it. Default output path matches the existing console's own convention: ~/.cache/agentm/telemetry/console.html.

Both modes are read-only: nothing here mutates the repo, the vault, or the GitHub project board. The board-drift section calls the read-only drift detector only (one gh issue list read) — never the comment-posting /report-board-drift cycle. The heat-policy section always runs in dry-run mode (never passes --apply).

What each section composes (and where to look if a section says "n/a")

• Health — scripts/status.py, which reads scripts/health/health_score.py's health-history ledger (the vault, when one resolves, else a device-local fallback — resolve_history_path()). Says "n/a" outside an agentm dev checkout, or if no scorecard has ever been recorded (run bash scripts/health/run-fast-tier.sh | python3 scripts/health/health_score.py --history once).
• Plans — scripts/queue_status_lite.py, the coordinator's read-only glance over every active _harness/ plan.
• Board drift — crickets' src/github-projects/scripts/check_project_sync.py (a sibling crickets checkout is required; the plugin's own .harness/project.json must resolve a github.repo). Says "n/a" if no crickets sibling is found, or reports the detector's own graceful-skip line if this repo isn't board-synced.
• Spend — scripts/runner/aggregator.py refreshes the SQLite rollup (best-effort; a missing crickets sibling means no refresh, not a failure) and scripts/health/observability_console.py reads it.
• Memory activity — inbox count (<vault>/personal/_inbox/*.md), a watchlist summary (harness/skills/memory/scripts/watchlist_review.py's own entry-listing function), incubator count (<vault>/_idea-incubator/*/), the five most-recently-modified curated entries under <vault>/personal/ (excluding the inbox/watchlist/archive staging areas), and the heat-based always-load decay report (recall.py heat-policy, dry-run). Says "n/a" if no vault resolves ($MEMORY_VAULT_PATH or plugins.obsidian-vault.vault_path).

Known scope limits (v1)

• No recall-trace substrate — "why did entry Y get recalled into this session" is out of scope for v1 (it needs machinery this skill does not build; see CONSOLIDATION-VERDICT.md Ruling 7).
• The Health/Plans/Spend/Board-drift sections are agentm-dev-checkout-aware: they read this repo's own scripts/health/, scripts/runner/, and a crickets sibling. On a bare downstream install (agentm consumed as a harness, not developed directly), those four sections report "n/a" and only Memory activity has real data — expected, not a bug, since those four surfaces are this repo's own dev-tooling and are not shipped to installs (see install.sh — nothing under scripts/health/, scripts/runner/, or scripts/control_plane/ is copied to a consumer project).