HermesCoderAgent

Branch, commit, push, and open PRs with humanized messages, then monitor GitHub Actions CI. Per-project autonomy gating; blocks direct pushes to the default branch and pushes with a dirty tree. Never auto-merges.

Dispatch patterns for Claude Code as the coding engine. Default harness. Implementation dispatches go through dispatch_coder.py (writes the dispatch receipt the commit gate requires).

Testing strategy, TDD enforcement, spec compliance, and regression checks.

Write implementation plans: bite-sized tasks for coding engine dispatch.

Automated fix loop for failed Quality/Reviewer checks. Parses failures, builds escalating prompts, dispatches through the active harness up to 3 times.

Run independent, file-disjoint plan tasks concurrently, each isolated in its own git worktree + branch. Collects per-task results; never auto-merges.

T-shirt sizing for tasks before dispatch. Runs heuristics and optionally LLM classification to determine task size (S/M/L/XL), recommend local vs cloud routing, and set tool budgets.

Make GitHub Issues the canonical backlog. Initialize the namespaced label schema, classify metadata, create/enrich/triage context-rich issues, and run the weekly grooming sweep. Opt-in per repo; autonomy-gated; never merges; groom may close stale/duplicate issues only behind the gate.

Capture a reusable lesson after an auto-heal struggle or debugger session, and inject relevant prior lessons into future dispatch prompts. Per-repo memory loop.

Enforced 4-phase debugging pipeline. Blocks source edits until bug is reproduced, traced, and hypothesis tested. Integrates with auto-healer for the fix phase.

CI/CD, deployment, infrastructure, and environment configuration.

Code review, PR management, and cross-concern quality synthesis.

System design, architecture decisions, dependency analysis for coding projects.

Dispatch patterns for Claude Code as the coding engine. Default harness.

Documentation, changelogs, API docs, and README maintenance.

Dispatch patterns for Antigravity CLI (agy) as the coding engine.

Pre-commit verification: security scan, quality gates, independent code review.

Task execution templates and coding engine dispatch patterns for coding tasks.

Security review, vulnerability scanning, and dependency audit for code changes.

Hardware-probe + sandboxed execution of vaulted tools. Network-isolated, read-only, resource-capped; never runs untrusted code on the host (#6, Phase 4).

Enforcement gate for all external prose. Strips AI slop patterns and calibrates to user voice before commits, PRs, docs, and chat summaries.

Route tasks to local Ollama models or cloud APIs based on triage output. Health-check aware with automatic cloud fallback.

On first touch of an unconfigured repo, interview the user about its permissions — backlog-as-GitHub-Issues, remote autonomy for PRs/pushes, and external skill discovery — then persist the answers to the repo's two config files so every later command honors them without re-asking. Allow skip (safe defaults: gated / no backlog / local-only).

Static + LLM code audit of a skill/tool source into FAIL/WARN/PASS. Never executes the code. The first guard in the dynamic ingestion pipeline (#6, Phase 4).

Discover task-relevant skills from a curated allowlist of trusted indexes, reputation-gate them, vet through the

End-to-end ingestion of a third-party skill/MCP source — fetch → quarantine → classify → audit → vault. A FAIL audit hard-blocks; Tier 2/3 needs --confirm (#6, Phase 4).

Classify a skill/tool source into a trust tier, hash it (SHA-256), and pin approved sources in an immutable local vault. Safe groundwork for the

Dispatch patterns for OpenCode as the coding engine.

Plan mode: write markdown plan to .hermes/plans/, no exec.

Throwaway experiments to validate an idea before build.

TDD: enforce RED-GREEN-REFACTOR in coding engine dispatches.