AI Slop Sweep

AI Slop Sweep

Overview

Detects AI-generated quality anti-patterns that violate ONEX platform invariants. These are patterns that slip through normal code review because they look syntactically valid but violate architectural contracts or CLAUDE.md invariants.

Announce at start: "I'm using the aislop-sweep skill."

Supported Repo List (default scan target)

Aislop scan is intentionally narrower than "all repos" to avoid signal noise from config-only, vendored, or non-Python repos:

AISLOP_REPOS = [
  "omniclaude", "omnibase_core", "omnibase_infra",
  "omnibase_spi", "omniintelligence", "omnimemory",
  "onex_change_control", "omnibase_compat"
]

Excluded by default: omnidash (Node.js), omniweb (PHP), omninode_infra (k8s YAML, no Python src/). Use --repos to override.

Autonomous execution: No Human Confirmation Gate. --dry-run is the only preview mechanism. Without it, proceed directly to ticket creation and/or fix dispatch.

CLI

/aislop-sweep                                   # Full scan all repos
/aislop-sweep --dry-run                        # Report only
/aislop-sweep --ticket                         # Create Linear tickets for findings
/aislop-sweep --checks phantom-callables,todo-fixme
/aislop-sweep --repos omniclaude,omnibase_core
/aislop-sweep --auto-fix                       # Fix trivial patterns

Path Exclusions

Applied to every scan and grep:

.git/          .venv/         node_modules/
__pycache__/   *.pyc          dist/          build/
docs/          examples/      fixtures/      _golden_path_validate/
migrations/    *.generated.*  vendored/

Check Categories + Conservative Triage Policy

Default action by severity tier:

Severity	Default action
CRITICAL	Always ticket (even in dry-run, emit as finding)
ERROR (HIGH confidence)	Ticket if --ticket set
WARNING	Report only — no ticket unless --ticket + --severity-threshold WARNING
INFO	Report only

Conservative first: Only prohibited-patterns, hardcoded-topics, and high-confidence phantom-callables create tickets by default. Context-sensitive checks (compat-shims, empty-impls, todo-fixme) are WARNING/INFO and require explicit --ticket --severity-threshold WARNING to create tickets.

Check	Pattern	Severity	Confidence rule
prohibited-patterns	ONEX_EVENT_BUS_TYPE=inmemory, OLLAMA_BASE_URL in .py/.sh	CRITICAL	HIGH always
hardcoded-topics	"onex\. string literal in .py outside contract.yaml / enum	ERROR	HIGH if in src/, MEDIUM if in tests/
phantom-callables	Executable-looking identifier() or call identifier in imperative skill .md context (not prose description, not examples); identifier absent from _lib/, _bin/, any .py under plugins/	ERROR	HIGH if confirmed missing (searched 3 locations); MEDIUM if only 1 location checked
compat-shims	# removed, # backwards.compat, _unused_ in non-test src	WARNING	MEDIUM
empty-impls	^\s*pass$ in src/ outside Abstract/Protocol/stub files	WARNING	MEDIUM
todo-fixme	TODO, FIXME, HACK in src/ (not tests/, not docs/)	WARNING	MEDIUM
todo-stale	TODO(OMN-XXXX) where referenced ticket is Done/Canceled in Linear	ERROR	HIGH (cross-referenced against Linear); falls back to WARNING if Linear MCP unavailable

ModelSweepFinding Schema

Every finding is a structured object:

{
  "repo":        str,      # e.g. "omniclaude"
  "path":        str,      # repo-relative path
  "line":        int,      # 0 if whole-file
  "check":       str,      # e.g. "prohibited-patterns"
  "message":     str,      # human-readable description
  "severity":    str,      # CRITICAL | ERROR | WARNING | INFO
  "confidence":  str,      # HIGH | MEDIUM | LOW
  "autofixable": bool,
  "ticketable":  bool,     # confidence=HIGH AND severity>=WARNING
}

Fingerprint: f"{repo}:{path}:{check}:{symbol_or_line_bucket}"

Execution Algorithm

1. PARSE arguments; resolve repo list and check set

2. SCAN (parallel, up to --max-parallel-repos):
   For each repo + each enabled check:
     Run grep pattern or static analysis
     Collect (file, line, pattern, severity) tuples
   Aggregate into findings[]

3. TRIAGE:
   Apply per-finding confidence + file-context rules:
   - prohibited-patterns in .py/.sh → CRITICAL, HIGH confidence always
   - hardcoded-topics in src/ → ERROR, HIGH; in tests/ → WARNING, MEDIUM
   - phantom-callables: verify identifier absent from _lib/, _bin/, any .py under plugins/
     → HIGH if confirmed missing; MEDIUM if path match is ambiguous
   - compat-shims: flag only in src/ (skip tests/, docs/, skill markdown)
   - empty-impls: skip *Abstract*, *Protocol*, *stub*, *__init__* files
   - todo-fixme: flag only in src/ (not tests/, not docs/, not skill markdown); severity=WARNING, confidence=MEDIUM
   - todo-stale: for each TODO(OMN-XXXX), query Linear MCP for ticket state;
     if state.type is COMPLETED or CANCELED → severity=ERROR, confidence=HIGH, ticketable=true
     if Linear MCP unavailable → downgrade to WARNING with explicit message:
     "Linear MCP unavailable -- todo-stale downgraded from ERROR to WARNING"
   Set ticketable=true when confidence=HIGH AND severity>=WARNING

4. IF no findings → emit ModelSkillResult(status=clean), exit

4a. FINGERPRINT + DEDUP:
   Compute fingerprints; load $ONEX_STATE_DIR/aislop-sweep/latest/findings.json
   Mark finding.new = fingerprint not in prior_fingerprints
   Save to $ONEX_STATE_DIR/aislop-sweep/<run_id>/findings.json

4b. GROUP findings:
   Group key = (repo, check_family, path)
   At most one ticket per group key unless severity=CRITICAL

5. IF --dry-run → print findings table by category and confidence tier, exit

6. IF --ticket:
   For each group: ticketable=true AND new=true AND severity>=threshold:
     Create Linear ticket: "aislop: <check_family> in <repo>:<path>"
     Parent: active sprint, label: aislop-sweep

7. IF --auto-fix:
   Auto-fix allowlist (extremely narrow):
   - Missing SPDX headers → stamp via `onex spdx fix`
   - `# removed` comment on blank line → safe to remove
   NOT auto-fixed: compat-shims with content, `_unused_` vars, empty `pass`
   Commit + PR per repo after confirming change via post-fix re-grep.

8. SUMMARY: Slack notification (best-effort)

9. EMIT ModelSkillResult

ModelSkillResult

{
  "skill": "aislop-sweep",
  "status": "clean | findings | partial | error",
  "run_id": "20260317-130000-b2c",
  "repos_scanned": 8,
  "total_findings": 23,
  "by_severity": {"CRITICAL": 1, "ERROR": 5, "WARNING": 12, "INFO": 5},
  "by_check": {
    "phantom-callables": 3,
    "compat-shims": 8,
    "prohibited-patterns": 1,
    "hardcoded-topics": 2,
    "todo-fixme": 7,
    "todo-stale": 2,
    "empty-impls": 2
  },
  "tickets_created": 6,
  "auto_fixed": 3
}

Status values:

• clean — zero findings
• findings — findings reported (tickets/fixes applied if requested)
• partial — some repos failed to scan
• error — scan failures prevented completion