Repo Scanner

Repo Scanner Specialist

Overview

A meta-specialist that exercises OSS Agent Lab's repository-intake shape. It derives a score from the literal slug and can copy the local skeleton when the synthetic score crosses the threshold.

Wraps jeremylongshore/oss-agent-lab.

The scanner does not contact GitHub. Structure flags, scores, and recommendations are derived from the repository string. Only scaffolding can write, by copying _template inside this checkout.

Prerequisites

• Use Python 3.11+ in a local OSS Agent Lab checkout and run pip install -e ..
• Provide a literal owner/repo slug and a safe snake_case target name.
• Obtain explicit user approval before a scaffold write and read the runtime contract.

Capabilities

• auto_scaffold: Materialise a new specialist directory from the _template skeleton when a repo's composite score reaches >= 80.
• repo_analysis: Simulate structural flags and infer likely capabilities from repo naming.
• specialist_generation: Orchestrate the full scan → score → scaffold pipeline and surface a unified result with provenance metadata.

Tools

Parameters

All parameters are passed via request.intent.parameters:

Instructions

1. Validate the literal owner/repo slug and run scan/score without claiming GitHub was queried.
2. Present the deterministic score, recommendation, proposed target path, and intended file set.
3. Ask for explicit approval before scaffold_specialist creates a directory.
4. After creation, report the exact path and files; the generated skeleton still requires review.

Examples

Python API

from agents.specialists.repo_scanner.agent import RepoScannerSpecialist
from oss_agent_lab.contracts import Intent, Query, SpecialistRequest

specialist = RepoScannerSpecialist()

request = SpecialistRequest(
    intent=Intent(
        action="scan",
        domain="meta",
        confidence=0.95,
        parameters={"repo": "openai/swarm"},
    ),
    query=Query(user_input="openai/swarm"),
    specialist_name="repo_scanner",
)

response = await specialist.execute(request)
print(response.result["score"]["action"])   # "auto_scaffold" | "evaluate" | "watch" | "skip"
print(response.result.get("scaffold"))      # None or {"status": "created", "path": ..., "files": [...]}

CLI

oss-lab run repo_scanner "openai/swarm"

Output shape

{
  "repo": "openai/swarm",
  "scan": {
    "repo": "openai/swarm",
    "name_suggestion": "swarm",
    "capabilities_detected": ["agent_orchestration"],
    "has_python": true,
    "has_tests": true,
    "has_readme": true,
    "recommendation": "auto_scaffold"
  },
  "score": {
    "repo": "openai/swarm",
    "estimated_score": 87.04,
    "action": "auto_scaffold",
    "signals": {
      "discovery": 32.1,
      "quality": 28.5,
      "durability": 26.44,
      "github_star_velocity": 0.87,
      "readme_quality": 0.91,
      "test_coverage": 0.75,
      "maintenance_activity": 0.60,
      "community_depth": 0.44
    }
  },
  "scaffold": {
    "status": "created",
    "path": "/home/jeremy/000-projects/oss-agent-lab/agents/specialists/swarm",
    "files": ["__init__.py", "agent.py", "SKILL.md", "tools.py"]
  }
}

Scoring Thresholds

Output

Read-only operations return simulated structural signals and scores. Scaffolding returns created only after copying the local template; if the template is absent it returns simulated and writes nothing. Neither status proves that the source repository exists or is suitable.

Error Handling

• Reject malformed repository slugs, unsafe names, and an existing target directory.
• Treat a missing template as no-write simulation, not successful creation.
• Never overwrite a specialist or broaden the destination beyond agents/specialists/<name>.

Resources

Wraps jeremylongshore/oss-agent-lab. See the runtime contract for scoring and write boundaries.