Skip to main content
ADLC FRAMEWORK38 AGENTS · TALENT BENCH7-PHASE LIFECYCLE100 ANTI-PATTERNS1 HITL / PHASE≥99.5% TARGETADLC v3.7.3FinOps · CloudOps · DevOps · AIOps · DataOps
LIVE · NZST
ADLC · Framework Command Centre
👤HITLHuman-in-the-Loop
🎯POProduct Owner
🏗️CACloud Architect
⚙️SPECSpecialists
BLOCKING · Zero Exceptions
§1 · AUTHORITY CHAIN (BLOCKING)
Source: CLAUDE.md 'Authority Chain (BLOCKING)'
👤
HITL
Human-in-the-Loop
Manager / T-Shape
All phases
🎯
PO
Product Owner
Business validation
Phase 1–2
🏗️
CA
Cloud Architect
Technical design
Phase 1–2
⚙️
SPEC
Specialists
38 execution agents
Phase 1–2 [A]
PHASE 1–2 [A] AUTONOMOUS
Think → Plan → Build · Agent-driven execution. PO+CA coordinate scope and design. Specialists execute under file-lock directives.
PHASE 3+ [HITL] APPROVAL REQUIRED
Test → Deploy → Monitor → Operate · HITL gates every phase transition. Agents prepare. Humans decide. Humans commit.
§2 · TALENT BENCH · 38 AGENTS
Source: CLAUDE.md Agent Team (38 agents)
§3 · 7-PHASE LIFECYCLE
Source: CLAUDE.md L56 · [A] = Autonomous · [HITL] = HITL Approval Required
THINK
[A]
Surface assumptions
PLAN
[A]
Design + scope
BUILD
[A]
Implement + test
TEST
[HITL]
Validation gate
DEPLOY
[HITL]
Production push
MONITOR
[HITL]
DORA + alerting
OPERATE
[HITL]
Continuous ops
[A] AutonomousPhases 1–3 · Agent-driven
[HITL] ApprovalPhases 4–7 · Human decides
§4 · 7 KARPATHY CODING DISCIPLINE RULES
Source: .claude/rules/engineering/coding-discipline.md
R1BLOCKS
Think Before Coding
Surface assumptions; ask don't guess; present alternatives before implementing.
HIDDEN_ASSUMPTION
R2BLOCKS
Simplicity First
Min code to solve stated problem. No Strategy/Factory for 1 implementation.
PREMATURE_ABSTRACTION
R3BLOCKS
Surgical Changes
Touch only what was requested. Every changed line traces to the request.
DRIVE_BY_REFACTORING
R4BLOCKS
Goal-Driven PDCA
Define success criteria in testable form before writing any code.
SUCCESS_CRITERIA_MISSING
R5BLOCKS
Subagent Context Preservation
Command files = sole delivery channel to subagent execution context. Never strip.
DOMAIN_CONTENT_STRIPPING
R6BLOCKS
RED-GREEN-REFACTOR
Write failing test FIRST. Min code to pass. Refactor under green only.
TESTING_THEATER
R7BLOCKS
Try Autonomous First
Attempt autonomous execution before deferring. Hook denial = only valid stop.
LAZY_DEFERRAL
§5 · ANTI-PATTERNS TOP-20 (of 100)
Source: .claude/rules/governance/anti-patterns-catalog.md
PATTERNSEVERITYDESCRIPTION
STANDALONE_EXECUTIONBLOCKINGSpecialist work without PO+CA approval
NATO_VIOLATIONBLOCKINGCompletion claims without evidence paths in tmp/
COORDINATION_WITHOUT_DELEGATIONBLOCKINGPO+CA ran but raw Edit/Write used instead of specialist Task
SCOPE_DRIFT_BYPASSRULES-LAYERPrior-scope coordination logs reused for new scope
TEXT_OUTPUT_BYPASSRULES-LAYERImplementation content in text without PO+CA coordination
FALSE_AS_IS_STATERULES-LAYERPlanning on fabricated baseline not cross-validated
KISS_MISINTERPRETATIONRULES-LAYERKISS applied to evidence richness instead of code complexity
NARROW_SEARCH_SCOPERULES-LAYERPer-account search when org-wide discovery tools exist
JSON_ONLY_NATORULES-LAYEREvidence JSON updated without updating HITL-consumable deliverable
AGENT_GIT_COMMIT_SUGGESTIONBLOCKINGAgent suggests or attempts git mutations — violates Principle I
TESTING_THEATERBLOCKINGHigh pass counts but mocks without assertions, inflated omit
RACE_CONDITION_SCORINGRULES-LAYERConcurrent scoring agents see different intermediate states
SCORING_THEATERBLOCKINGRe-scoring unchanged artifacts — PDCA count increments, nothing changes
NO_ESTIMATED_COUNTSRULES-LAYERNumeric claims by estimation, not verification command output
AWS_PROFILE_SEMANTIC_MISMATCHRULES-LAYERWrong profile for API — each profile has exactly ONE correct purpose
MOCK_CIRCULAR_VALIDATIONBLOCKINGTest asserts hardcoded mock value that code also returns — circular
PARALLEL_AGENT_SCOPE_BLEEDRULES-LAYERParallel agents create/modify files outside assigned scope
PLAN_DELIVERY_MISMATCHRULES-LAYERSprint plan claims features not in shipped code
LAZY_DEFERRALRULES-LAYERClaiming deferred without attempting autonomous execution first
HIDDEN_ASSUMPTIONRULES-LAYERAgent picks interpretation silently without surfacing assumption
§6 · LIVE DORA METRICS + COMPONENT INVENTORY
Source: /data/dora.csv (live) · /data/components.csv (live) · Framework rows only
FRAMEWORK DORA
PDCA
HITL 1
Deploy Frequency
3/sprint
target: ≥2/sprint10/10
Lead Time
<2 days
target: <2 days9/10
Change Failure Rate
0%
target: <5%10/10
MTTR
N/A
target: <1 hour
COMPONENT INVENTORYBrowse Marketplace →
38
Agents
144
Commands
130
Skills
51
Hooks
22
Rules
GOVERNANCE ACCURACY TARGET
≥99.5%
PDCA composite · target ≥95%
§7 · ROI · ROADMAP · ADOPTION SIGNALS
Sources: CLAUDE.md L90 · VERSION · docs/static/data/stories.csv · git log · session-telemetry.jsonl
FRAMEWORK ROI
cited sources only — no orphan numbers
AI-augmented velocity
vs single-agent baseline
3–5×
CLAUDE.md L90
Sprint time-box
Scrum Guide 2020 ≤1 month
1 week
CLAUDE.md L90
Anti-pattern catalog
patterns with hook enforcement
100
anti-patterns-catalog.md
DORA band — Deploy Freq
DF ≥2/sprint · CFR 0%
Elite
/data/dora.csv (live)
Governance target
PDCA composite pass rate
≥99.5%
/data/dora.csv (live)
COMPONENTS (live · /data/components.csv)
38Agents
144Commands
130Skills
51Hooks
22Rules
2026–2027 ROADMAP
planned milestones — source: VERSION + CLAUDE.md
2026-Q1
ADLC v3.7 stable · 38-agent Talent Bench
source: VERSION → 3.7.3
2026-Q2
SPM XC-→AIO- namespace · JIRA 3-way sync
source: CLAUDE.md sprint model
2026-Q3
BMC v2 · 5 products × 9 blocks · ≥99.5% gate
source: planned
2026-Q4
Sovereign AI · APRA CPS 234 evidence pack
source: planned
2027-Q1
ADLC GA · AWS APN + Atlassian Marketplace
source: planned
ADOPTION SIGNALS
verified at build time — commands cited per claim
Framework stories
cited: docs/static/data/stories.csv
21
wc -l stories.csv
Completed stories
cited: stories.csv status col
9
grep -c completed
JIRA rows (all products)
cited: wc -l jira/*.csv
150
6 CSV projections
Commits since 2026-04-01
cited: git log --since
82
adlc-framework
Telemetry events
cited: wc -l transparency/
464
session-telemetry.jsonl
Agent invocations
task adlc:trace:summary
TBD pending audit