Eight pillar readiness map

Eight-Pillar Readiness Map

Overview

The dual-graph architecture is the structural framework; the eight pillars are the implementation roadmap — the specific capabilities that turn the framework into a working system. Each pillar emerged from a recurring production failure pattern, and the pillars are layered, not independent: knowledge representation (Ch3) must come first because every other pillar depends on the knowledge graph; memory (Ch4) builds on it; reasoning and planning (Ch5) require both; tool orchestration (Ch6) requires all of those; and the final two, self-evolution (Ch7) and optimization (Ch8), depend on the complete architecture being in place.

This skill takes a system's declared capability per pillar (present / partial / missing) and produces a readiness map:

• a readiness score (weighted across the eight pillars),
• dependency violations — a higher pillar claimed present while a lower one it requires is missing (the chapter says this is structurally unsound),
• unresolved Chapter-1 flaws — mapping remaining gaps back to the five flaws per Table 2-1 (relationship blindness, context amnesia, temporal ignorance, reasoning paralysis, tool chaos),
• the next pillar to build (always the earliest incomplete layer, because building higher first creates a violation).

The DevOps agent's Chapter-2 initial state is the worked example: knowledge representation only partial (the data exists in logs and configs but is not a graph), everything else missing, all five flaws open. It is a "representation problem, not a technology problem."

When to Use

• Auditing an agentic-graph system's production readiness
• Planning the build order for a new agent (which pillar next?)
• Checking that a claimed capability is not resting on a missing foundation
• Tracking progress across the book's parts (each part adds a pillar-layer)

Phrases that should invoke this skill: "map the eight pillars", "which pillar should we build next", "is this agent production-ready", "pillar readiness", "what flaws are still unsolved", "eight-pillar audit".

When NOT to Use

• Building a single pillar. Each pillar has its own chapter and skills (knowledge-representation/, memory/, reasoning-planning/, tool-orchestration/, self-evolution/). This skill maps; it does not build.
• Routing a request. Vertical-vs-horizontal routing is dual-graph-router.
• A generic org/AI maturity model. The eight pillars are specific to the dual-graph architecture; do not repurpose this for unrelated capability grids.
• Scoring the QUALITY of a pillar. This tracks presence/partial/missing, not how good a present pillar is — use the pillar's own evaluation skills for that.

Process

Step	Input	Action	Output	Verification
1	capability dict {pillar: present|partial|missing}	lib.assess(caps)	ReadinessReport	unstated pillars default to missing; invalid status raises ValueError
2	statuses	lib.dependency_violations(statuses)	list of violations	a present/partial pillar with a missing required lower pillar is flagged
3	statuses	lib.unresolved_flaws(statuses)	list of open flaws	a flaw is solved only if a solving pillar is fully present
4	statuses	lib.next_pillar(statuses)	pillar key or None	the earliest not-present pillar in layer order
5	—	lib.initial_state()	the DevOps Chapter-2 starting capabilities	KR partial; all others missing

Rationalizations

Agent rationalization	Documented rebuttal
"We built reasoning first because it is the interesting part — skip the graph for now."	The chapter is explicit: "Knowledge representation must come first because every other pillar depends on the knowledge graph." Reasoning present while KR missing is a dependency violation — the reasoning has nothing grounded to reason over.
"Self-evolution is the goal, let's jump to Chapter 7."	Self-evolution "depends on the complete architecture being in place." It reads the observation record produced by the harness that the earlier pillars build. Claimed present on an incomplete stack, it is a violation, not a capability.
"Tool orchestration solves tool chaos, so once we have it we are done with flaws."	tool_chaos also needs structured_output at the tool/node boundaries (the impedance-mismatch pillar). A present tool_orchestration with missing structured_output leaves tool_chaos only partially addressed.
"Partial on a pillar is basically present."	Partial means the flaw it targets is only partially addressed. The initial DevOps state has partial KR (data exists, unstructured) and still leaves relationship_blindness open — partial is not solved.
"We can build the pillars in any order our team prefers."	The pillars are layered by dependency, not preference. Building out of order produces the dependency violations this skill exists to catch; the recommended next pillar is always the earliest incomplete layer.

Red Flags

• A high readiness score with dependency violations. The weighted score can look healthy while the layering is broken — always read the violations list, not just the percentage.
• Every flaw shows unsolved but pillars are marked present. The status-to-flaw mapping drifted, or pillars were marked present without their dependencies — cross-check with the violations list.
• next_pillar skips ahead of a missing lower pillar. Impossible by construction; if you see it, PILLAR_ORDER or depends_on was edited incorrectly.
• self_evolution or optimization shown as solving a flaw. They map to production viability, not to any of the five flaws; a flaw mapping to them is a corruption of Table 2-1.
• CLI --help exits non-zero. SKILL.md / CLI mismatch; the multi-harness invariant is broken.

Non-Negotiable Verification

1. Run the benchmark battery.

   python cli.py benchmark
   

   Confirm eight layered pillars, five mapped flaws, the initial state recommending knowledge_representation, the mid state solving the first four flaws while leaving tool_chaos open, and the violation state flagging reasoning + self_evolution as resting on a missing foundation.

2. Inspect the initial state visually.

   python cli.py assess --system initial
   

   Confirm KR is partial, all others missing, all five flaws unsolved, and the roadmap starts at knowledge_representation.

3. Confirm the layering catches a violation.

   python cli.py assess --system violation
   

   Confirm reasoning and self_evolution are flagged as claimed-present-but- foundation-missing.

4. JSON output round-trips.

   python cli.py assess --system mid --json | python -c "import json,sys; json.load(sys.stdin)"
   

   No exception means the CLI is harness-portable.

Security Posture

• Prompt injection. Capability statuses are a small closed vocabulary (present/partial/missing) validated on input; any other value raises ValueError rather than being interpreted. Pillar keys are validated against the fixed eight; unknown keys raise. There is no free-text execution surface.
• Data exfiltration. No network calls, no file writes outside the read-only bundled sample-systems.json. The --json report goes to stdout; the caller owns downstream piping.
• Advisory only. This skill emits a report; it never edits the audited system, installs a pillar, or grants a capability. A readiness score is not an authorization — the dependency-violation output is the safety signal (do not ship a system whose claimed pillars rest on missing foundations).

Source Attribution

Distilled from Agentic GraphRAG (O'Reilly, by Anthony Alcaraz and Sam Julien), Chapter 2 — Agentic Graph Architecture Foundations:

• "The Eight Pillars of Agentic Graph Architecture" — the eight pillars, their layering, and why each exists (a recurring production failure pattern).
• Table 2-1 "Mapping the five flaws to the eight pillars" — relationship blindness → knowledge representation; context amnesia + temporal ignorance → memory; reasoning paralysis → reasoning + planning; tool chaos → tool orchestration + structured output.
• "The Initial State" and "The Transformation Arc" — the DevOps agent's fragmented starting point and the pillar-per-part progression.

This Reviewer-pattern skill audits the whole architecture; the per-pillar skills in the sibling chapter folders build the capabilities it maps.