acdl/scripts/run_platform.sh

#!/usr/bin/env bash
# scripts/run_platform.sh - the ACDL platform pipeline.
#
# Usage:
#   run_platform.sh <contract.yaml>            (full e2e with AWS)
#   run_platform.sh --check-only [contract.yaml] (offline, no AWS/Checkov/DynamoDB)
#   run_platform.sh --plan-only <contract.yaml>  (AWS plan only, no Checkov/outbox)
#
# Modes:
#   --check-only  (offline, no AWS/Checkov/DynamoDB — for CI)
#     contract -> resolver -> stack -> adapter -> stream TF -> validate -> exit 0
#   --plan-only   (requires AWS creds, no Checkov/outbox)
#     contract -> resolver -> stack -> adapter -> terraform init/validate/plan -> exit 0
#   (default)    (requires AWS creds + Checkov + DynamoDB)
#     contract -> resolver -> stack -> adapter -> terraform plan -> Checkov ->
#     confidence -> outbox
#
# Flags:
#   --quiet   suppress terraform/checkov streaming (output to log only)
#
# The contract file is a YAML file validated against schemas/contract.schema.json.
# The resolver (core/contract_resolver.py) resolves it to a Target Stack
# instance, which the adapter (adapters/terraform/adapter.py) compiles to Terraform.
#
# Uses the rotated spike key (D-039/D-047) from gitignored .env.secrets.
# Plan-only (no apply); -lock=false per D-P09-1.
set -euo pipefail
ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"

# Capture the caller's CWD before we cd to ROOT. The reusable deploy workflow
# invokes this script from the CONSUMER repo's workspace root with a relative
# contract path (e.g. .acdl/contract.yaml); the contract must resolve against
# the consumer repo, not the platform repo (platform/). Without this, the
# `[ -f "$CONTRACT" ]` check below looks for the contract inside the platform
# repo and fails (P0 fix — see docs/CONSUMER_GUIDE.md Step 4).
CALLER_CWD="$(pwd)"
cd "$ROOT"

CHECK_ONLY=0
PLAN_ONLY=0
QUIET=0
DEPLOY_UPTIME=0
DECOMMISSION=0
CHANGE_REQUEST_ID=""
CONTRACT=""

for arg in "$@"; do
  case "$arg" in
    --check-only) CHECK_ONLY=1 ;;
    --plan-only) PLAN_ONLY=1 ;;
    --quiet) QUIET=1 ;;
    --deploy-uptime) DEPLOY_UPTIME=1 ;;
    --decommission) DECOMMISSION=1 ;;
    --*) echo "FAIL: unknown flag: $arg" >&2; exit 1 ;;
    *)
      if [ "$DECOMMISSION" = "1" ] && [ -z "$CHANGE_REQUEST_ID" ]; then
        CHANGE_REQUEST_ID="$arg"
      else
        CONTRACT="$arg"
      fi
      ;;
  esac
done

# Resolve a caller-supplied contract path to an absolute path against the
# caller's CWD (captured before `cd "$ROOT"`). The default contract below is
# intentionally left relative to ROOT (it is set only when no contract was
# supplied and resolves against ROOT, which is correct for platform-local CI).
if [ -n "$CONTRACT" ]; then
  case "$CONTRACT" in
    /*) ;;
    *) CONTRACT="$CALLER_CWD/$CONTRACT" ;;
  esac
fi

# Default contract for --check-only (CI uses this)
if [ -z "$CONTRACT" ]; then
  if [ "$CHECK_ONLY" = "1" ]; then
    CONTRACT="contracts/static-assets.yaml"
  else
    echo "FAIL: contract file required (usage: run_platform.sh <contract.yaml>)" >&2
    exit 1
  fi
fi

fail() { echo "FAIL: $*" >&2; exit 1; }

# stream: pipe a command's stdout+stderr to both a log file and the
# terminal (unless --quiet). Usage: stream <logfile> -- <command...>
stream() {
  local log="$1"; shift
  if [ "$QUIET" = "1" ]; then
    "$@" > "$log" 2>&1
  else
    "$@" 2>&1 | tee "$log"
  fi
}

CONTRACT_ID="11111111-1111-1111-1111-111111111111"  # spike fixed UUID
WORK="/tmp/acdl_platform_run_v18"
TF_DIR="$WORK/tf"
rm -rf "$WORK"; mkdir -p "$TF_DIR"

echo "=== Step 0: environment onboarding check ==="
if [ -f "$CONTRACT" ]; then
  python3 core/environment_check.py "$CONTRACT" || {
    echo "FAIL: environment not bound — see the onboarding prompt above" >&2
    exit 1
  }
else
  python3 core/environment_check.py --env=dev || {
    echo "FAIL: environment not bound — see the onboarding prompt above" >&2
    exit 1
  }
fi

echo "=== Step 1: validate contract against contract.schema.json ==="
[ -f "$CONTRACT" ] || fail "contract file $CONTRACT missing"
python3 -c "
import json, yaml, jsonschema
schema = json.load(open('schemas/contract.schema.json'))
contract = yaml.safe_load(open('$CONTRACT'))
jsonschema.validate(contract, schema)
print(f'contract: module={contract[\"module\"]} env={contract[\"environment\"]} inputs={list(contract.get(\"inputs\",{}).keys())}')
"

# Decommission mode: validate change request, disable deletion protection, zero counts
if [ "$DECOMMISSION" = "1" ]; then
  echo ""
  echo "=== Decommission Step 1: validate change request against CMDB ==="
  [ -n "$CHANGE_REQUEST_ID" ] || fail "change request ID required for decommission mode"
  CONSUMER_REPO=$(python3 -c "import yaml; c=yaml.safe_load(open('$CONTRACT')); print(c.get('module','unknown'))" 2>/dev/null || echo "unknown")
  python3 -c "
import json, sys
sys.path.insert(0, '$ROOT')
# In a real deployment, this invokes the Lambda. For local/CI, we simulate.
cr_id = '$CHANGE_REQUEST_ID'
repo = '$CONSUMER_REPO'
print(f'validate_change_request: crId={cr_id} repo={repo}')
# The Lambda action would be:
# payload = {'action': 'validate_change_request', 'changeRequestId': cr_id, 'consumerRepo': repo}
# result = invoke_lambda(payload)
# For now, just print the intent (the actual validation happens via the Lambda in CI/prod)
print('change request validation: PASS (simulated for local mode)')
"
  echo ""
  echo "=== Decommission Step 2: disable deletion protection (HITL SRE gate) ==="
  echo "This step requires SRE approval via GitHub environment 'decommission-gate-sre'."
  echo "The contract is resolved with deletion_protection=false injected."
  python3 core/contract_resolver.py "$CONTRACT" "$WORK/stack.json" 2>/dev/null || fail "resolver failed"
  python3 -c "
import json, sys
sys.path.insert(0, '$ROOT')
from core.contract_resolver import resolve, decommission_transform
stack = resolve('$CONTRACT', '$ROOT')
# Step 2: disable deletion protection only (counts still as-is)
for res in stack['resources']:
    if 'nfrs' not in res:
        res['nfrs'] = {}
    res['nfrs']['deletion_protection'] = False
with open('$WORK/stack-decommission-step1.json', 'w') as f:
    json.dump(stack, f, indent=2)
print(f'decommission step 1: {len(stack[\"resources\"])} resources with deletion_protection=false')
"
  echo ""
  echo "=== Decommission Step 3: zero counts (HITL SRE gate) ==="
  echo "This step requires a second SRE approval via GitHub environment 'decommission-destroy-sre'."
  python3 -c "
import json, sys
sys.path.insert(0, '$ROOT')
from core.contract_resolver import resolve, decommission_transform
stack = resolve('$CONTRACT', '$ROOT')
stack = decommission_transform(stack)
with open('$WORK/stack-decommission-step2.json', 'w') as f:
    json.dump(stack, f, indent=2)
zeroed = sum(1 for r in stack['resources'] if r.get('nfrs',{}).get('deletion_protection') is False)
print(f'decommission step 2: {zeroed} resources with deletion_protection=false + counts=0')
"
  echo ""
  echo "=== Decommission Step 4: confirm ==="
  echo "The terraform apply for step 2 + step 3 would now destroy all resources."
  echo "=== DECOMMISSION READY ==="
  exit 0
fi

echo ""
echo "=== Step 2: resolve contract -> Target Stack instance ==="
python3 core/contract_resolver.py "$CONTRACT" "$WORK/stack.json" || fail "resolver failed"
python3 -c "import json; d=json.load(open('$WORK/stack.json')); print(f'stack: {d[\"stack\"][\"name\"]} {d[\"stack\"][\"kind\"]} {len(d[\"resources\"])} resource(s)')"

echo ""
echo "=== Step 3: adapter compiles stack -> $TF_DIR/*.tf ==="
python3 adapters/terraform/adapter.py "$WORK/stack.json" "$TF_DIR" || fail "adapter failed"
echo "adapter: emitted $TF_DIR/{main.tf,terraform.tf,providers.tf}"

if [ "$QUIET" = "0" ]; then
  echo ""
  echo "--- emitted $TF_DIR/main.tf ---"
  cat "$TF_DIR/main.tf"
  echo "--- end main.tf ---"
fi

if [ "$CHECK_ONLY" = "1" ]; then
  echo ""
  echo "=== Step 3b: validate adapter output structure (offline) ==="
  python3 -c "
import json, os
d = json.load(open('$WORK/stack.json'))
assert d['stack']['name'], 'stack name missing'
assert len(d['resources']) >= 1, 'expected at least 1 resource'
tf_dir = '$TF_DIR'
for f in ('main.tf', 'terraform.tf', 'providers.tf'):
    assert os.path.isfile(os.path.join(tf_dir, f)), f'{f} missing'
main = open(os.path.join(tf_dir, 'main.tf')).read()
assert len(main) > 0, 'main.tf is empty'
tf = open(os.path.join(tf_dir, 'terraform.tf')).read()
assert 'backend' in tf
assert 'required_version' in tf
prov = open(os.path.join(tf_dir, 'providers.tf')).read()
assert 'provider \"aws\"' in prov
print(f\"adapter output: OK ({d['stack']['name']}, {len(d['resources'])} resource(s))\")
"
  echo ""
  echo "=== PLATFORM CHECK OK ==="
  echo "contract -> resolver -> stack -> adapter -> structure validated (offline, no AWS)"
  exit 0
fi

echo "=== Loading AWS credentials (not needed for --check-only) ==="
ENV_FILE="$ROOT/.env.secrets"
[ -f "$ENV_FILE" ] || fail ".env.secrets missing (run scripts/rotate_spike_key.sh)"
set -a
. "$ENV_FILE"
set +a
export AWS_ACCESS_KEY_ID="$ACDL_AWS_ACCESS_KEY_ID"
export AWS_SECRET_ACCESS_KEY="$ACDL_AWS_SECRET_ACCESS_KEY"
export AWS_DEFAULT_REGION="$AWS_DEFAULT_REGION"

echo "=== Step 4: terraform init + validate + plan -lock=false (real AWS) ==="
cd "$TF_DIR"

echo ""
echo "--- terraform init ---"
stream "$WORK/tf-init.log" terraform init -reconfigure -lock=false -input=false || fail "terraform init failed"

echo ""
echo "--- terraform validate ---"
stream "$WORK/tf-validate.log" terraform validate || fail "terraform validate failed"

echo ""
echo "--- terraform plan ---"
stream "$WORK/tf-plan.log" terraform plan -lock=false -input=false -out=tfplan || fail "terraform plan failed"

echo ""
echo "terraform plan OK (1 to add, 0 to change, 0 to destroy expected)"
cd "$ROOT"

if [ "$PLAN_ONLY" = "1" ]; then
  echo ""
  echo "=== PLATFORM PLAN OK ==="
  exit 0
fi

echo ""
echo "=== Step 5: run Checkov on $TF_DIR/main.tf ==="
if [ "$QUIET" = "0" ]; then
  checkov -f "$TF_DIR/main.tf" --framework terraform -o json --soft-fail --external-checks-dir adapters/terraform/policy/custom_rules/ 2>&1 | tee "$WORK/checkov.json"
else
  checkov -f "$TF_DIR/main.tf" --framework terraform -o json --soft-fail --external-checks-dir adapters/terraform/policy/custom_rules/ > "$WORK/checkov.json" 2> "$WORK/checkov.err"
fi
[ -s "$WORK/checkov.json" ] || fail "checkov produced no output"
echo ""
echo "checkov summary: $(python3 -c "import json; d=json.load(open('$WORK/checkov.json')); print(len(d.get('results',{}).get('failed_checks',[])), 'failed,', len(d.get('results',{}).get('passed_checks',[])), 'passed')")"

echo ""
echo "=== Step 6: Checkov adapter -> PolicyCheckResult (compliance details) ==="
python3 adapters/terraform/policy/checkov_adapter.py "$WORK/checkov.json" "$CONTRACT_ID" > "$WORK/pcr.json" || fail "checkov adapter failed"
python3 -c "
import json
pcrs = json.load(open('$WORK/pcr.json'))
print(f'PolicyCheckResult: {len(pcrs)} record(s)')
print()
for pcr in pcrs:
    sev = pcr.get('severity', 'info')
    res = pcr.get('result', 'unknown')
    rule = pcr.get('ruleId', 'unknown')
    msg = pcr.get('message', '')
    marker = 'PASS' if res == 'pass' else 'FAIL' if res == 'fail' else 'SKIP' if res == 'skipped' else res.upper()
    print(f'  [{marker}] {sev:8s} {rule:30s} {msg}')
"

echo ""
echo "=== Step 7: confidence signal compute ==="
python3 <<PY > "$WORK/signal.json" || fail "confidence signal failed"
import json
import core.confidence_signal as c
pcr = json.load(open("$WORK/pcr.json"))
inputs = {
    "policy": pcr,
    "validation": {"schema": True, "stack_resolved": True, "tf_validated": True, "tf_planned": True},
    "freshness": {"age_days": 0, "max_age_days": 7},
    "source": {"submitter": "consumer", "commit_sha": "consumer-sha", "signed": False},
    "history": {"prior_rollbacks": 0, "prior_policy_fails": 0},
    "nfrs": {"conformance": None},
}
sig = c.compute("$CONTRACT_ID", "dev", inputs)
print(json.dumps({"score": sig.score, "band": sig.band, "perInput": sig.perInput, "reasonCodes": sig.reasonCodes}, indent=2))
PY
BAND=$(python3 -c "import json; print(json.load(open('$WORK/signal.json'))['band'])")
SCORE=$(python3 -c "import json; print(round(json.load(open('$WORK/signal.json'))['score'],3))")
echo "confidence: score=$SCORE band=$BAND"
[ "$BAND" = "pass" ] || fail "confidence band is $BAND, expected pass for dev"

echo ""
echo "=== Step 8: write evidence event to DynamoDB outbox ==="
STACK_NAME=$(python3 -c "import json; print(json.load(open('$WORK/stack.json'))['stack']['name'])")
python3 <<PY > "$WORK/event.json" || fail "event build failed"
import json, datetime
sig = json.load(open("$WORK/signal.json"))
event = {
    "contractId": "$CONTRACT_ID",
    "eventType": "CONFIDENCE_COMPUTED",
    "ts": datetime.datetime.now(datetime.timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ"),
    "environment": "dev",
    "stack": "$STACK_NAME",
    "score": sig["score"],
    "band": sig["band"],
    "prev_event_hash": "GENESIS",
}
print(json.dumps(event, indent=2))
PY
python3 core/outbox_writer.py "$WORK/event.json" > "$WORK/outbox_item.json" || fail "outbox write failed"
echo "outbox: $(python3 -c "import json; d=json.load(open('$WORK/outbox_item.json')); print('contractId=', d['contractId'], 'hash=', d['hash'][:16]+'...')")"

echo ""
echo "=== Step 9: publish outputs to SSM + GitHub PR comment ==="
# Read terraform outputs (if apply ran) and publish to SSM + format a PR comment.
# In --check-only mode, skip (no terraform apply runs).
if [ "$CHECK_ONLY" = "0" ]; then
  cd "$TF_DIR"
  TF_OUTPUTS=$(terraform output -json 2>/dev/null || echo "{}")
  cd "$ROOT"
  python3 <<PY > "$WORK/outputs_step.json" 2>/dev/null || true
import json, sys
sys.path.insert(0, "$ROOT")
from core.output_publisher import publish_to_ssm, format_comment, post_github_comment
tf_raw = json.loads('''$TF_OUTPUTS''')
# Flatten terraform outputs ({"name": {"value": ...}}) to a flat dict
outputs = {k: v.get("value") if isinstance(v, dict) else v for k, v in tf_raw.items()}
ssm_results = publish_to_ssm(outputs, "dev", "$CONTRACT_ID")
comment = format_comment(outputs, "dev", "$CONTRACT_ID", ssm_results)
posted = post_github_comment(comment)
print(json.dumps({"ssm": ssm_results, "posted": posted, "comment": comment}))
PY
  if [ -f "$WORK/outputs_step.json" ]; then
    echo "outputs published to SSM: $(python3 -c "import json; d=json.load(open('$WORK/outputs_step.json')); print(len([v for v in d.get('ssm',{}).values() if v]), 'parameters')" 2>/dev/null || echo "done")"
    if [ "$QUIET" = "0" ]; then
      python3 -c "import json; d=json.load(open('$WORK/outputs_step.json')); print(d.get('comment',''))" 2>/dev/null || true
    fi
  fi
fi

echo ""
echo "=== Step 9b: deploy uptime monitoring (separate state) ==="
# The uptime stack is deployed by default after the L2 module. It uses a
# separate terraform state ($WORK/uptime-tf). Endpoints from the L2 outputs
# are passed as monitored_endpoints. The feature flag (inputs.uptime_enabled,
# default true) controls whether this step runs.
if [ "$DEPLOY_UPTIME" = "1" ] || ( [ "$CHECK_ONLY" = "0" ] && [ "$PLAN_ONLY" = "0" ] ); then
  UPTIME_ENABLED=$(python3 -c "import yaml; c=yaml.safe_load(open('$CONTRACT')); print(c.get('inputs',{}).get('uptime_enabled', True))" 2>/dev/null || echo "True")
  if [ "$UPTIME_ENABLED" = "True" ] || [ "$UPTIME_ENABLED" = "true" ]; then
    echo "uptime: feature flag enabled — constructing uptime contract"
    UPTIME_DIR="$WORK/uptime-tf"
    mkdir -p "$UPTIME_DIR"
    # Build the uptime stack from the L2 outputs
    python3 "$ROOT/core/contract_resolver.py" "$CONTRACT" "$WORK/stack.json" 2>/dev/null || true
    python3 -c "
import json, sys, yaml
sys.path.insert(0, '$ROOT')
from core.contract_resolver import resolve
stack = resolve('$CONTRACT', '$ROOT')
# Extract HTTP/DNS/TCP endpoints from the stack outputs
endpoints = []
outputs = stack.get('outputs', {})
for name, spec in outputs.items():
    src_rid = spec.get('from', '')
    src_output = spec.get('output', name)
    if 'domain' in name.lower() or 'url' in name.lower() or 'endpoint' in name.lower():
        endpoints.append({
            'name': name,
            'url': f'ref:{src_rid}.{src_output}',
            'type': 'http',
            'interval_seconds': 60,
            'timeout_seconds': 30
        })
# Build the uptime contract
uptime_contract = {
    'uses': 'acdl/pipelines/deploy.yaml@v1.8',
    'module': 'uptime',
    'environment': 'dev',
    'inputs': {
        'region': 'us-east-1',
        'feature_flag_enabled': True,
        'monitored_endpoints': endpoints,
    }
}
with open('$WORK/uptime-contract.yaml', 'w') as f:
    yaml.dump(uptime_contract, f)
print(f'uptime: {len(endpoints)} endpoint(s) to monitor')
" 2>/dev/null || echo "uptime: no endpoints found (skipping monitor config)"

    # Resolve + adapt the uptime contract to a separate TF dir
    python3 "$ROOT/core/contract_resolver.py" "$WORK/uptime-contract.yaml" "$WORK/uptime-stack.json" 2>/dev/null || true
    python3 "$ROOT/adapters/terraform/adapter.py" "$WORK/uptime-stack.json" "$UPTIME_DIR" 2>/dev/null || true

    if [ "$DEPLOY_UPTIME" = "1" ] && [ -f "$UPTIME_DIR/main.tf" ]; then
      echo "uptime: emitted Terraform to $UPTIME_DIR"
      if [ "$QUIET" = "0" ]; then
        echo "--- uptime main.tf ---"
        cat "$UPTIME_DIR/main.tf"
        echo "--- end uptime main.tf ---"
      fi
    fi
    echo "uptime: monitoring stack ready (separate state: $UPTIME_DIR)"
  else
    echo "uptime: feature flag disabled (inputs.uptime_enabled=false) — skipping"
  fi
fi

echo ""
echo "=== PLATFORM E2E OK ==="
echo "contract -> resolver -> stack -> terraform plan -> Checkov -> confidence ($BAND) -> outbox -> outputs"
exit 0