feat(agent): SolFoundry Bounty Agent for auto-submitting to T1 bounties

[akuraposo]

Summary

Autonomous Claude agent (700 lines, zero hard deps beyond requests) that:

1. Discovers Tier-1 bounties matching its capabilities
2. Generates solution code/templates
3. Forks the repo, commits, pushes
4. Creates a properly formatted PR (with Closes #N + wallet)

What this PR adds

• automaton/agents/sol_foundry_agent.py (18.4 KB, 560 lines) — full agent implementation

Acceptance criteria (from #845)

• Automatic bounty discovery based on agent capabilities
• Solution implementation (template-based)
• Automated PR submission with proper formatting

Architecture

1. Capability definition

@dataclass
class Capability:
    name: str
    languages: list[str]   # Python, TypeScript, Solidity, Markdown, Bash
    domains: list[str]     # backend, frontend, docs, agent
    tools: list[str]       # git, github-api, foundry
    min_reward: int
    max_reward: int

2. Bounty parsing

Extracts from GitHub issue body:

• Reward: regex ([\d,]+)\s*K?\s*\$FNDRY or \$([\d,]+)
• Tier: regex Tier[-:\s]+(T[123])
• Domains: from labels (backend, frontend, docs, etc.)

3. Scoring algorithm (0-1)

• Reward in range: +0.3
• Tier T1 (open race): +0.2
• Domain overlap: +0.3 (Jaccard-weighted)
• Language match: +0.1
• Clear acceptance criteria: +0.1

Filter: only T1 with score >= 0.3

4. Solution generation

Type-aware templates:

• Doc bounties -> markdown file
• Component bounties -> React .tsx + test
• Workflow bounties -> n8n JSON
• Generic -> Python template

5. Submission pipeline

fork_repo() -> commit_and_push() -> create_pr()
  -> Comment /claim on issue
  -> Generate files based on bounty type
  -> git checkout -B agent/bounty-N (from main, clean branch)
  -> git add + commit + push to fork
  -> POST /repos/{upstream}/pulls

CLI usage

# Discover + auto-submit top 3 bounties
python3 sol_foundry_agent.py --max-bounties 3

# Target specific bounty
python3 sol_foundry_agent.py --bounty 830

# Test without submitting
python3 sol_foundry_agent.py --dry-run

Features

• Zero hard deps — only requests
• Multi-platform — works on any GitHub repo with bounty label
• Configurable — capability, min reward, max bounties, target repos
• Race-aware — only targets T1 (open race, no reputation gates)
• Dry-run mode — test discovery without submitting

Closes #845

Wallet: JDuuZK8brNFergaNqWo4pfEYuv5oCLUGxqxcFDMYAcwb