Feature Configuration

!bash ~/.claude/hooks/sweetclaude/record-event.sh skill_invoked "sweetclaude:_features" 2>/dev/null || true

Feature Configuration

Step 1: Read current state

python3 - << 'PY'
import yaml, os

sc_path = '.sweetclaude/state/sweetclaude.yaml'
try:
    with open(sc_path) as f:
        d = yaml.safe_load(f) or {}
except:
    d = {}

features = d.get('features', {})

def get_status(key):
    v = features.get(key)
    if isinstance(v, dict):
        return v.get('status', 'not_configured')
    return 'not_configured'

statuses = {k: get_status(k) for k in [
    'product_milestones', 'product_backlog', 'product_personas',
    'product_stories', 'document_corpus', 'usage_tracking', 'behavioral_regression',
]}

# usage_tracking: ground truth is the metrics config file, not just the yaml status
try:
    mc = yaml.safe_load(open('.sweetclaude/metrics/config.yaml'))
    if mc.get('enabled', False):
        statuses['usage_tracking'] = 'active'
except:
    pass

enabled = sum(1 for s in statuses.values() if s == 'active')
total = len(statuses)
all_unconfigured = all(s not in ('active', 'declined') for s in statuses.values())

print(f"ENABLED:{enabled}")
print(f"TOTAL:{total}")
print(f"FIRST_TIME:{str(all_unconfigured).upper()}")
for k, s in statuses.items():
    print(f"{k}:{s}")
PY

Step 2: Present features and collect choices

Use this table for all feature labels and descriptions:

Key	Label	Description
product_milestones	Product milestones	Milestone tracking — set concrete targets and see progress toward them
product_backlog	Product backlog	Running list of everything to build — keeps ideas from falling through the cracks
product_personas	User personas	Define your target users — clearer personas make every product decision easier
product_stories	User stories	Turn ideas into testable acceptance criteria — the direct input to TDD and code
document_corpus	Document corpus	Connect your docs for automatic reference — SweetClaude searches them so you don't re-explain context
usage_tracking	Usage tracking	Local metrics: skill usage, phase gate outcomes, TDD enforcement — stored in .sweetclaude/metrics/, committed with the project, no external services
behavioral_regression	Behavioral regression	Verify SweetClaude follows framework rules after model updates — catches regressions in Claude's behavior

If FIRST_TIME is TRUE:

"SweetClaude has optional features — here's what's available. Select the ones you want enabled:"

Otherwise:

"This project has {ENABLED} of {TOTAL} features enabled. Select all the features you want on — including any already enabled:"

Use AskUserQuestion (multiSelect: true) with all 7 features as options:

• label: the feature label — append " ✓" if the feature's current status is active
• description: the feature description from the table above

Step 3: Write decisions

Compute the final state from the user's selections:

• Selected → active
• Not selected → declined

Track which features are newly enabled (selected AND were not previously active) — those need onboarding in Step 4.

Write all decisions in one atomic update:

python3 - .sweetclaude/state/sweetclaude.yaml KEY STATUS [KEY STATUS ...] << 'PY'
import sys, yaml, tempfile, os
from datetime import datetime, timezone

path = sys.argv[1]
pairs = sys.argv[2:]
now = datetime.now(timezone.utc).isoformat(timespec='seconds')

with open(path) as f:
    d = yaml.safe_load(f) or {}

features = d.setdefault('features', {})
it = iter(pairs)
for key, status in zip(it, it):
    if not isinstance(features.get(key), dict):
        features[key] = {}
    features[key].update({'status': status, 'decided_at': now})

with tempfile.NamedTemporaryFile('w', dir=os.path.dirname(path), suffix='.tmp', delete=False) as tmp:
    yaml.dump(d, tmp, default_flow_style=False, allow_unicode=True, sort_keys=False)
    tmp_name = tmp.name
os.replace(tmp_name, path)
print('OK')
PY

Step 4: Onboard newly-enabled features

For each feature that is newly enabled (was not previously active), invoke its onboarding action. Do not move to the next feature until the current one either completes or explicitly signals it is paused.

Feature key	Action
product_milestones	invoke sweetclaude:product-milestones with argument onboard
product_backlog	invoke sweetclaude:project-backlog with argument onboard
product_personas	invoke sweetclaude:user-personas with argument onboard
product_stories	invoke sweetclaude:product-user-stories with argument onboard
document_corpus	invoke sweetclaude:document-corpus with argument onboard
usage_tracking	invoke sweetclaude:usage on
behavioral_regression	invoke sweetclaude:behavioral-regression

Handling a paused onboard: Some skills (notably document_corpus) run multi-step pipelines requiring active user review at each stage. If a skill's onboard output contains a pause marker (e.g. "⏸ … paused … resume with …"), surface that message verbatim, then output: "Moving on — I'll set up the remaining features and you can resume corpus setup at any time." Then continue to the next feature. Do NOT treat a pause marker as onboard completion.

If no features are newly enabled, say: "Feature configuration saved." and stop.