feat(examples): Slack auto-fix agent — monitors channel and auto-creates PRs

[manan164]

What this does

Adds 91_slack_autofix_agent.py — an autonomous engineering agent that monitors a Slack channel for bug reports and automatically investigates, fixes, and opens a GitHub PR.

Architecture

slack_autofix_pipeline (SEQUENTIAL)
├── issue_reader       — reads Slack channel, extracts bug description
├── code_investigator  — searches codebase, finds root cause
├── code_fixer         — applies minimal fix to the relevant files
└── pr_creator         — creates branch, commits, pushes, opens PR, replies in Slack

Tools

Tool	What it does
fetch_slack_bug_reports	Slack API — reads new messages from the bug channel
search_codebase	grep -rn over the repo to find relevant code
read_file / write_file	Read and write files in the repo
run_git_command	Run arbitrary git commands
create_branch_and_commit	Create branch + stage + commit
push_branch	Push to remote origin
create_github_pr	gh pr create — opens the PR
mark_message_processed	State file — prevents re-processing the same bug
post_slack_reply	Reply in the Slack thread with the PR link

Usage

# Required env vars
export SLACK_BOT_TOKEN=xoxb-...
export SLACK_CHANNEL_ID=C...
export GITHUB_TOKEN=ghp_...
export GITHUB_REPO=agentspan-ai/agentspan
export REPO_PATH=/path/to/agentspan
export AGENTSPAN_SERVER_URL=http://localhost:6767/api
export AGENTSPAN_LLM_MODEL=anthropic/claude-opus-4-6

# Run once
uv run python examples/91_slack_autofix_agent.py

# Dry-run (investigate only, no writes or PRs)
uv run python examples/91_slack_autofix_agent.py --dry-run

# Continuous loop (poll every 5 minutes)
uv run python examples/91_slack_autofix_agent.py --loop --interval 300

Why Agentspan for this

• Durable — agent run survives restarts; if the fixer crashes mid-commit, Conductor retries
• Observable — every tool call, LLM decision, file write logged in Conductor UI
• Auditable — full trace of what the agent read, changed, and why — before the PR is merged
• Sequential guarantee — code_fixer never runs before code_investigator completes

🤖 Generated with Claude Code

[bradyyie]

LGTM — Well-documented Slack auto-fix agent example. Good demonstration of AgentSpan's sequential strategy for a real-world use case (Slack monitoring -> code investigation -> fix -> PR creation). Clear README, practical usage instructions, and proper env var handling.

Dismissing approval — need to properly reproduce bugs before approving. Re-reviewing with hands-on QA.