247a225d54db2012dc2f17434f63b663c551499a
4 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Jon Chery
|
e00393adfd |
decision(P03): clarification — L2/contract/evidence/hash-chain schemas
---ci---
phase: 3
milestone: v1.0
status: clarify
decisions:
- id: D-020
decision: L2 manifest.yaml schema with l1s: list of {name, inputs: map}
rationale: REQ-04 says L2 composes L1s; REQ-05 caps depth at 5 (L2->L1 is depth 1)
confidence: 0.90
alternatives: [graph-of-L1s with edges, single flat list of L1 names with no inputs]
- id: D-021
decision: contract.yaml schema with stack, inputs, optional public-ingress: bool
rationale: REQ-07 cites public-ingress:true as the forbidden key; REQ-08 keys off policy pass/fail
confidence: 0.90
alternatives: [separate policy file, contract-free intent]
- id: D-022
decision: mock_executor.sh writes state.json with l2, l1s array (name+applied+exit_code), contract
rationale: REQ-06 says writes state.json but does not specify shape
confidence: 0.85
alternatives: [per-L1 state files, no state file]
- id: D-023
decision: evidence_writer.py appends to audit.json (JSON array); each event has seq, ts, stage, event, prev_hash, hash (SHA-256 of canonical JSON); genesis has prev_hash=GENESIS, seq=0
rationale: D-005 mandates hash-chained ledger; canonical JSON for deterministic hashing
confidence: 0.90
alternatives: [signed commits only, append-only log file]
- id: D-024
decision: confidence_signal.py prints {score, reason} JSON to stdout; exit 0 always
rationale: REQ-08 literal: base 0.90, drops to 0.40, gate >= 0.50
confidence: 0.95
alternatives: [exit non-zero on fail, separate side-channel file]
- id: D-025
decision: policy_checker.py exits 1 with stdout POLICY_VIOLATION:PUBLIC_INGRESS on public-ingress:true; else exit 0 with stdout POLICY_PASS
rationale: REQ-07 literal
confidence: 0.99
alternatives: []
- id: D-026
decision: l3b_agent_stub.py reads issue body from argv[1] or stdin, applies D-008 keyword map, writes contract.yaml to stdout (or -o <path>)
rationale: D-008 + Act 3 example; L3B must produce same contract format as L3A
confidence: 0.90
alternatives: [fixed-output contract per stack, no -o flag]
---/ci---
Phase 03 has real ambiguities (no schema was specified for L2 manifests,
contract.yaml, audit.json events, or the L3B output format). Seven
decisions logged: D-020 L2 schema, D-021 contract.yaml schema, D-022
state.json shape, D-023 audit.json event + hash chain, D-024
confidence_signal output, D-025 policy_checker output, D-026 l3b_agent
I/O contract.
|
||
 grimacing
|
00d0043866 |
ship: phase-02 l1-modules (v1.0.2)
Squash merge of phase/02-l1-modules into milestone/v1.0-initial; 8 L1 stub modules created; verify_phase02.sh green. |
||
|
Jon Chery
|
e5f5604319 |
decision(P01): clarification — project mode, pages fallback, env gates
---ci---
phase: 1
milestone: v1.0
status: clarify
decisions:
- id: D-011
decision: Single-project mode explicitly enforced via config.json mode=single
rationale: run.md Step 0 reads projects[] length as multi-project trigger; explicit flag disambiguates from D-010
confidence: 0.95
alternatives: [move acdl into .ciagent/acdl/ subdirectory and adopt multi-project paths]
- id: D-012
decision: Gitea has no native Pages; serve acdl-evidence via raw file URLs + CORS note
rationale: Research confirms Gitea has no [pages] section; raw URLs work without server config
confidence: 0.85
alternatives: [sidecar static server, external Pages host]
- id: D-013
decision: QA/Prod gates modeled as workflow_dispatch approval inputs (D-004 fallback) instead of Gitea environments
rationale: Research confirms Gitea ignores environment: blocks and exposes no environments API
confidence: 0.90
alternatives: [external approval bot, drop approval gates entirely]
- id: D-014
decision: Cross-repo triggering uses workflow_dispatch API from inside a step (no repository_dispatch)
rationale: Gitea Actions does not support repository_dispatch
confidence: 0.85
alternatives: [push-based trigger with a sentinel file, polling]
- id: D-015
decision: New repos acdl-contracts and acdl-evidence use default_branch=main with auto_init=true
rationale: Matches Gitea DEFAULT_BRANCH=main; required for the default branch to exist before any push
confidence: 0.95
alternatives: [use milestone/v1.0-initial as default_branch]
- id: D-016
decision: Pages placeholder for Phase 01 is a minimal HTML stub; full timeline UI deferred to Phase 05
rationale: Phase 01 success criterion is the URL returns 200 with placeholder; full UI is Phase 05
confidence: 0.90
alternatives: [build a minimal timeline now]
---/ci---
Clarifications accepted at full autonomy per clarify.md Step 4. Three requirements
re-stated: REQ-09 (new repos default_branch=main), REQ-10 Pages (raw-URL 200
substitute), REQ-10 environments (workflow_dispatch inputs + qa/prod branches).
See .ciagent/REQUIREMENTS.md Clarifications table.
|
||
|
Jon Chery
|
e3416f8e77 |
docs(init): initialize Agentic Cloud Delivery Platform (5 phases)
---ci---
phase: 0
milestone: v1.0
status: specify
decisions:
- id: D-001
decision: Use Gitea org continuous-intelligence for all ACDL repos
rationale: User-specified target org; already exists at git.cloudinit.dev
confidence: 0.95
alternatives: [new dedicated demo org]
- id: D-002
decision: Map "GitHub Actions" to Gitea Actions (act_runner) using same workflow YAML
rationale: Environment is Gitea; syntax-compatible with act_runner
confidence: 0.85
alternatives: [migrate to GitHub.com, raw shell scripts]
- id: D-003
decision: Collapse acdl-platform into the existing empty acdl repo
rationale: acdl already exists at org root; avoids a 4th repo
confidence: 0.90
alternatives: [create separate acdl-platform repo]
- id: D-004
decision: Use Gitea environment blocks + required reviewers for QA/Prod; fallback to manual workflow_dispatch with approval input
rationale: Spec mandates approval gates; forge supports environment protection
confidence: 0.80
alternatives: [external approval bot, no approval gates]
- id: D-005
decision: Hash-chained ledger (prev_hash + own hash via SHA-256 of canonical JSON) for evidence; declared demonstrative not adversarially secure
rationale: Spec asks for simple JSON; chain gives visible tamper-evidence
confidence: 0.85
alternatives: [signed commits only, full Merkle tree]
- id: D-006
decision: Confidence gate threshold = 0.50 exactly (base 0.90, fail drops to 0.40)
rationale: Explicit in spec
confidence: 0.99
alternatives: []
- id: D-007
decision: Each mock_apply.sh echoes "[L1: <name>] applying..." + "OK", sleeps 1s, exits 0
rationale: Spec literal; uniformity aids timeline parsing
confidence: 0.95
alternatives: [randomized sleep durations]
- id: D-008
decision: L3B keyword->stack mapping: gas/price/ingest/data-lake->commodity-price-feed; invoice/billing->invoice-service; analytics/historical/query->energy-analytics-api; regulatory/compliance/reporting/trading->regulatory-reporting; fallback->invoice-service
rationale: Mirrors the 4 L2 modules + Act 3 example issue text
confidence: 0.85
alternatives: [single default stack, ML classifier (forbidden by no-AI constraint)]
- id: D-009
decision: Init milestone = v1.0, branch milestone/v1.0-initial
rationale: init.md Step 5 mandates milestone/v1.0-initial
confidence: 0.99
alternatives: []
- id: D-010
decision: Single-project mode for the acdl checkout; ---ci--- blocks omit project field
rationale: User chose standalone single-project; no other projects in this checkout
confidence: 0.90
alternatives: [register acdl in /root multi-project config]
---/ci---
Specification: 30-minute executive demo of the Agentic Cloud Delivery Platform proving automatic, safe, audited infra delivery via local stubs on Gitea Actions (no cloud, no AI APIs).
Requirements: [3 repos under continuous-intelligence org, 8 L1 stub modules, 4 L2 composed modules mirroring S&P Global Energy use cases, 5 core scripts (mock_executor.sh, policy_checker.py, confidence_signal.py, evidence_writer.py, l3b_agent_stub.py), reusable Dev->QA->Prod pipeline with approval gates, issue-triggered L3B agent, Pages evidence timeline UI, 4 scripted demo acts]
Constraints: [local Linux OS, Gitea Actions + Environments, no AWS/GCP/Azure, no external LLM APIs, state in flat JSON/artifacts, EKS Fargate + serverless primitives (no VPC module), L1 single-purpose substrate-agnostic max-depth-1, L2 max-depth-5]
Out of scope: [real cloud provisioning, real LLM inference, production-grade infrastructure, adversarial tamper-proofing of evidence]
|