observe

name	observe
description	Single-pass network health observation. Checks API liveness, namespace freshness, drift, computes Network Health Score.
allowed-tools	["Bash","Read","Grep","Glob","Write","Edit"]
user-invocable	true
agent	Explore

Observe — Single-Pass Network Health Check

Purpose

Run a single observation pass across the entire constructs network. Produce structured JSONL observations and compute the Network Health Score. This is the atomic unit — /patrol calls this repeatedly, /report synthesizes its output.

Invocation

/observe                    # Full network pass
/observe --quick            # API-only (skip GitHub checks)

Workflow

Step 1: Load Baseline

Read the last observation from grimoires/gecko/observations.jsonl (if exists). Extract the previous health_score for ratchet comparison.

Step 2: Check API Liveness

# Health endpoints
curl -s https://api.constructs.network/v1/health | jq .
curl -s https://api.constructs.network/v1/health/ready | jq .

# Registry state
curl -s https://api.constructs.network/v1/constructs | jq '.data | length'

Record: api_status (healthy/degraded/down), registered_count, response_time_ms.

Step 3: Check Namespace Freshness

# List all construct-* repos under 0xHoneyJar
gh api orgs/0xHoneyJar/repos --paginate -q '.[] | select(.name | startswith("construct-")) | {name, pushed_at, archived}'

For each repo, record:

• slug: repo name minus construct- prefix
• last_push: ISO timestamp
• days_stale: days since last push
• archived: boolean

Flag constructs with days_stale > 30 as STALE. Flag days_stale > 90 as ABANDONED.

Step 4: Check Category Distribution

Query the API or parse construct.yaml files for domain[0] to determine category assignments. Count constructs per category across the 8 canonical categories:

• development, security, design, documentation, operations, infrastructure, research, straylight

Flag categories with 0 constructs as EMPTY_CATEGORY.

Step 5: Check Identity-Reality Drift (if not --quick)

For each construct in the namespace:

1. Read identity/persona.yaml — extract cognitiveFrame.archetype and voice.personality_markers
2. Read skills/*/SKILL.md — extract actual methodology steps
3. Compare: does the skill methodology reflect the persona's claimed expertise?
4. Score drift: ALIGNED (0), MINOR_DRIFT (1-2 mismatches), SIGNIFICANT_DRIFT (3+)

This step requires cloning or reading repos via gh api. For speed, use:

gh api repos/0xHoneyJar/construct-{slug}/contents/identity/persona.yaml -q .content | base64 -d
gh api repos/0xHoneyJar/construct-{slug}/contents/construct.yaml -q .content | base64 -d

Step 6: Compute Network Health Score

Weighted composite (0-100):

Sub-Signal	Weight	Scoring
API Liveness	0.20	healthy=100, degraded=50, down=0
Version Freshness	0.25	avg(100 - min(days_stale, 100)) across all constructs
Category Coverage	0.15	(filled_categories / 8) * 100
Identity-Reality Drift	0.20	avg(aligned=100, minor=70, significant=30) across all
Composition Density	0.10	(constructs_with_compose_with / total) * 100
Verification Flow	0.10	(non_unverified / total) * 100

Step 7: Emit Observation

Append to grimoires/gecko/observations.jsonl:

{
  "timestamp": "2026-03-12T00:00:00Z",
  "health_score": 72,
  "health_delta": -3,
  "api_status": "healthy",
  "registered_count": 15,
  "namespace_count": 16,
  "stale_constructs": ["webgl-particles"],
  "empty_categories": ["infrastructure"],
  "drift_detected": [{"slug": "dynamic-auth", "level": "MINOR_DRIFT"}],
  "anomalies": []
}

Step 8: Surface Findings

If health_delta < 0 (health degraded), surface the contributing sub-signals as findings. If health_delta >= 0, report quietly — the ratchet held.

Output format (to stdout):

gecko | health: 72 (-3) | api: healthy | stale: 1 | drift: 1 | categories: 7/8

If anomalies exist, list them. Otherwise, one line is enough.

Outputs

Path	Description
grimoires/gecko/observations.jsonl	Append-only observation log

Constraints

• Never modify construct source files
• Never extrapolate — if you can't measure it, don't score it
• API failures are degraded, not anomaly — the network survives without the registry
• Identity-reality drift is a signal, not a judgment — flag it, don't prescribe fixes