agentic-issue-solver

name	agentic-issue-solver
description	Use when the user wants one automated GitHub issue-solving pass against the current repository, wants to run the centralized issue solver, or wants to check whether the installed issue-solver skill is out of date

Agentic Issue Solver

Run exactly one centralized issue-solving workstream against the current repository.

This skill is intended for Codex installations where the skill directory is a symlink into a cloned checkout of shinaoka/agentic-issue-solver.

First Step: Check Installation Freshness

At the start of every invocation, run:

bash "${CODEX_HOME:-$HOME/.codex}/skills/agentic-issue-solver/scripts/check-installation.sh"

Interpret the result:

• status=ok: continue
• status=update-available: tell the user an update exists and show the exact git pull --ff-only command
• status=detached-copy: tell the user the skill is not running from a git checkout and recommend reinstalling via the documented clone-plus-symlink flow
• remote_status=unavailable: mention that update status could not be confirmed, then continue if the task is otherwise unblocked

Do not auto-update. Prompt the user to update, but continue when the request is still actionable on the installed version.

Scope

• One solver run only
• No built-in outer loop
• No long-lived campaign manager
• No assumption that target repositories vendor these scripts

One run may still address multiple issues when that produces a smaller, cleaner, lower-friction PR than splitting the work artificially.

Target Repository Inputs

Before invoking the solver, inspect the target repository:

• AGENTS.md
• ai/AGENTIC_ISSUE_SOLVER.md if present
• README.md when more context is needed

Repository-specific instructions may be natural language. Respect documented repository procedures for verification, PR creation, merge, CI monitoring, and closeout before using fallback behavior.

Choosing Backend

• If the user explicitly names a backend, use it
• Otherwise default to codex
• Do not ask the user to choose a model
• Let the selected backend CLI use its own default model and reasoning effort

Running the Solver

From the target repository root, or with an explicit path, run:

bash "${CODEX_HOME:-$HOME/.codex}/skills/agentic-issue-solver/scripts/run-solver.sh" \
  --target-repo /path/to/target-repo

Examples:

bash "${CODEX_HOME:-$HOME/.codex}/skills/agentic-issue-solver/scripts/run-solver.sh" \
  --target-repo . --backend codex

bash "${CODEX_HOME:-$HOME/.codex}/skills/agentic-issue-solver/scripts/run-solver.sh" \
  --target-repo . --backend claude

Result Handling

The solver prints a final single-line JSON summary. Use that summary to tell the user what happened:

• fixed
• closed_stale
• commented_no_fix
• no_actionable_issue
• failed

The result payload uses an issues array so a single workstream can report one or many handled issues.

If the user wants repeated runs or continuous PR/CI monitoring, that is outer orchestration. Run multiple single passes rather than inventing an internal loop.