system-check · diff
git:20260707.2a2e857 to git:20260726.7992efa
57 added, 37 removed. Audit A to A.
---
name: system-check
agent: none
- description: Full-stack smoke-test covering Somatic v0.4/v0.5 layer AND Cognitive v0.7 runtime. Skills, daemons, state files, runtime files, capsules, mirror discipline, evals, daemon error log. Reports green/red per path.
+ description: Read-only installation smoke test for skills, daemons, operational memory, cognitive runtime, nightly freshness, terminal evidence, and unique route ownership.
allowed-tools: Bash, Read
user-invocable: true
triggers:
- system check
- - somatic check
- smoke test
- check wiring
- is everything wired
- - audit somatic
- - cognitive check
+ - audit install
- runtime check
+ - nightly health
- /system-check
---
# /system-check
- Run `bash daemons/system-check.sh` from the vault root.
+ Run from the aigent-OS root:
- Covers the full stack — Somatic v0.4/v0.5 AND Cognitive v0.7.
+ ```bash
+ bash daemons/system-check.sh
+ ```
- ## Somatic layer (v0.4/v0.5)
+ The script is read-only. It prints one line per check, a final
+ `SUMMARY: N PASS / N FAIL / N INFO`, and exits `1` when any check fails.
- - **Skills (8):** body-check, digest, context-capsule, caddy-mute, system-check, sweep-now, capsule-compact, agent-fitness
- - **Daemons (8):** caddy.sh, memory-capture.sh, sync-usage.sh, log-token-usage.sh, system-check.sh, compute-heat.js, capsule-compact.py, agent-fitness-extract.py
- - **State files (7):** BODY_STATE.json (schema keys), HEAT_INDEX.json (JSON + mtime <7d), CADDY_MUTES.json (if present), MEMORY_CANDIDATES.md (table present), HESTIA_SWEEP_LOG.md, AGENT_FITNESS.md (header row), capsules dir (non-empty)
- - **Capsules health:** all capsules have valid frontmatter (status enum, capsule_id)
- - **Mirror discipline:** vault `.claude/skills/` and `daemons/` match public aigent-OS
+ ## Path resolution
- ## Cognitive layer (v0.7)
+ Framework code resolves from `AIGENT_ROOT`, defaulting to the parent directory
+ of `daemons/system-check.sh`.
- - **Skills (9):** status, reconcile, dream, meta-improve, skill-recall, skill-hunt, context-capsule, pause, resume
- - **Runtime daemon:** `daemons/runtime/update-active-state.py` — exists + parses
- - **Runtime state files (8):**
- - `ACTIVE_STATE.json` — valid JSON, `mode` in [stabilization, build, expansion, recovery]
- - `SELF_MODEL.json` — `capabilities` and `limitations` arrays non-empty
- - `GOAL_STACK.json` — `active_goals` array exists
- - `BELIEF_STATE.jsonl` — valid JSONL, last entry has `confidence` field
- - `LESSONS.jsonl` — valid JSONL
- - `PROCEDURES.jsonl` — valid JSONL
- - `STATE_EVENTS.jsonl` — valid JSONL
- - `DREAM_LOG.md` — exists; FAIL if last modified > 7 days ago
- - **Evals directory:** `evals/` exists and contains at least one file
+ Operational memory resolves in this order:
- ## Daemon error log
+ 1. `AIGENT_STATE_HOME_DIR/vault/memory` when the state-home diversion is set.
+ 2. `AIGENT_STATE_HOME_DIR/memory` as a compatibility fallback.
+ 3. `AIGENT_ROOT/vault/memory` for a normal installation.
+ 4. `AIGENT_ROOT/memory` only when the canonical vault path does not exist.
- Tail of `memory/.daemon-errors.log` — count + top 5 entries surfaced as INFO.
+ This is the same shape used by the nightly controller. Tests should set
+ `AIGENT_ROOT`, `AIGENT_STATE_HOME_DIR`, and date-sensitive environment values
+ explicitly so ambient state cannot change the result.
- Exits 0 if all green, 1 if any FAIL.
+ ## Checks
- ## When to run
+ - Required top-level skills exist, including the nightly close-parity skills.
+ - Shell and Python daemons parse.
+ - Every nightly daemon exists and passes `node --check`.
+ - The memory-heat daemon parses.
+ - Operational state files have their required schemas, including the generic
+ sweep log and dispatch-fitness table.
+ - Capsules have valid frontmatter, identity, and status.
+ - Canonical cognitive JSON and JSONL inputs exist and parse. Missing canonical
+ input is a failure, not an empty-success case.
+ - `DREAM_LOG.md` freshness comes from its newest valid dated header, never its
+ file modification time.
+ - The newest `NIGHTLY_LOG.md` pass is current and has complete terminal
+ evidence. A terminal `status: FAIL` remains red even when every checkpoint
+ row is present.
+ - `/nightly` resolves through the unique `/nightly-close-parity` route.
+ - The eval directory contains at least one supported file.
+ - The local daemon error log is surfaced as INFO when it contains entries.
- - On demand when something feels off
- - After shipping any Somatic or Cognitive version update
- - Before closing a session where runtime state was mutated
- - After `/dream` or `/reconcile` to confirm state files updated cleanly
+ ## Date configuration
- ## What it does NOT do
+ All date-sensitive checks use one timezone and one cutoff:
- - Does not fix anything. Read-only audit.
- - Does not validate semantic correctness — only that wired things parse and exist.
- - Does not run heavy computation. Should complete in <5 seconds.
+ ```bash
+ export AIGENT_NIGHTLY_TIME_ZONE="America/Los_Angeles"
+ export AIGENT_NIGHTLY_CUTOFF_HOUR="4"
+ bash daemons/system-check.sh
+ ```
+
+ `America/Los_Angeles` and hour `4` are the defaults. Tests may pin the current
+ instant with `AIGENT_SYSTEM_CHECK_NOW`.
+
+ ## What it does not do
+
+ - It does not repair files or resolve alerts.
+ - It invokes the route checker with alerts disabled.
+ - It invokes the watchdog in check-only mode, so `/system-check` never appends
+ to the nightly alert ledger.
+ - It does not inspect a message bus, task board, browser profile, or private
+ machine configuration.