@ghx-dev/core@0.1.2

Patch Changes

• d2b4892: SOLID improvements: error handling, dead code removal, and testability.

  • fix mapErrorToCode to use word-boundary regex for HTTP status codes and reorder checks to prevent keyword collisions
  • fix preflightCheck to return ADAPTER_UNSUPPORTED for missing gh CLI instead of VALIDATION
  • remove 7 unimplemented capability IDs from GraphqlCapabilityId union and update YAML cards to prefer CLI
  • separate internal sentinel params (__wasDraft, __effectiveRerunMode) from normalizeCliData params into dedicated context arg
  • consolidate duplicate AJV instances into shared ajv-instance.ts
  • export OperationCard type from public API
  • remove dead code: 66 unused task stubs, empty command scaffolds, stub formatter, unused ExecutionMetrics interface

@ghx-dev/core@0.1.1

Patch Changes

• 68a9cad: Benchmark updates:

  • Add a config-driven benchmark suite runner workflow with paired mode execution, progress events,
    fixture setup/cleanup commands, and reporting/gating orchestration.
  • Expand benchmark scenario coverage with seeded fixture bindings for issues, PRs, workflows,
    releases, and project-v2 flows, including fixture auth/bootstrap tooling.
  • Harden fixture resolution and seeding behavior (safer manifest binding paths, check-run id
    handling, stricter CLI input validation), and improve dashboard rendering/progress semantics.
  • Increase benchmark validation quality with broader schema/assertion checks and significantly
    expanded unit-test coverage across CLI, fixture, runner, and reporting modules.

  Core updates:

  • Extend CLI capability adapter coverage and handling for workflow id/input normalization and
    project-v2 item list normalization edge cases.

• 74d4a61: Core updates:

  • Externalize ghx setup skill text into a bundled SKILL.md asset, copy skill assets into dist,
    and load setup content from packaged asset paths at runtime.
  • Standardize setup install target to .agents/skills/ghx/SKILL.md (user and project scopes), and
    align setup docs/tests to the same canonical path.
  • Improve setup command robustness by handling ENOENT checks consistently, awaiting skill-file reads
    in the asset loader path, and adding/refreshing unit and e2e tests for setup install/verify and
    asset error scenarios.
  • Align setup usage guidance across core docs/readmes (--scope-based usage).

  Benchmark updates:

  • Fix scenario schema record typing for zod v4 compatibility.

  Repo tooling/docs updates included in this branch:

  • Update pre-commit typecheck hook environment configuration for consistent Nx behavior.
  • Add a dedicated setup command guide and synchronize onboarding references.

• 2179f98: Benchmark updates:

  • Refactor suite execution by splitting mode-instruction, preflight, and prompt-rendering logic
    into focused runner modules.
  • Align benchmark mode instructions and ghx preflight behavior, including stricter preflight
    checks for router readiness and fixture usage.
  • Improve prompt rendering and output schema handling, including pageInfo compatibility.
  • Add and update unit coverage for suite-runner, preflight, mode instruction, and prompt renderer
    paths.

  Core updates:

  • Adjust run-path routing behavior and CLI integration to align with benchmark runner updates.
  • Refresh CLI/engine tests around run command flow and repository view behavior.

@ghx-dev/core@0.1.0

Minor Changes

• eaba8ea: # First Public Release

  ghx is a typed GitHub execution router for AI agents. One capability interface. Deterministic routing. Normalized results. No more agents fumbling through gh CLI docs on every run.

  The Problem

  When agents interact with GitHub, they waste tokens re-discovering API surfaces, parsing inconsistent outputs, and handling errors ad hoc. Common PR operations that should be straightforward become multi-step research projects.

  What ghx Delivers

  Every capability returns a stable envelope: { ok, data, error, meta }. Route selection, retries, fallbacks, and input/output validation are handled automatically through schema-driven operation cards.

  Benchmarked Results (agent_direct vs ghx)

  Across 27 runs on common PR operations:

  Metric	Improvement
  Active tokens	-37% fewer tokens consumed
  Latency	-32% faster end-to-end
  Tool calls	-33% fewer tool invocations
  Success rate	100% for both (zero regressions)

  Agents using ghx skip the research phase entirely: no manual route discovery, no output guessing, no wasted reasoning tokens.

  66 Capabilities Shipped

  Repository and issues -- repo.view, issue.view, issue.list, issue.comments.list, issue.create, issue.update, issue.close, issue.reopen, issue.delete, issue.labels.update, issue.assignees.update, issue.milestone.set, issue.comments.create, issue.linked_prs.list, issue.relations.get, issue.parent.set, issue.parent.remove, issue.blocked_by.add, issue.blocked_by.remove

  PR review and merge -- pr.view, pr.list, pr.comments.list, pr.reviews.list, pr.diff.list_files, pr.status.checks, pr.checks.get_failed, pr.mergeability.view, pr.comment.reply, pr.comment.resolve, pr.comment.unresolve, pr.ready_for_review.set, pr.review.submit_approve, pr.review.submit_request_changes, pr.review.submit_comment, pr.merge.execute, pr.checks.rerun_failed, pr.checks.rerun_all, pr.reviewers.request, pr.assignees.update, pr.branch.update

  CI diagnostics -- check_run.annotations.list, workflow_runs.list, workflow_run.jobs.list, workflow_job.logs.get, workflow_job.logs.analyze

  Release and delivery -- release.list, release.get, release.create_draft, release.update, release.publish_draft, workflow_dispatch.run, workflow_run.rerun_failed, workflow_run.rerun_all, workflow_run.cancel, workflow_run.get, workflow_run.artifacts.list, workflow.list, workflow.get

  Projects v2 and repo metadata -- project_v2.org.get, project_v2.user.get, project_v2.fields.list, project_v2.items.list, project_v2.item.add_issue, project_v2.item.field.update, repo.labels.list, repo.issue_types.list

  Core Features

  • Deterministic routing -- each capability declares a preferred route and ordered fallbacks; the engine follows the plan without guesswork
  • Runtime validation -- AJV-powered input/output schema validation from operation cards catches errors before they reach GitHub
  • Structured error taxonomy -- canonical error codes (AUTH, NOT_FOUND, RATE_LIMIT, VALIDATION, ...) with retryable flags for predictable automation behavior
  • Attempt metadata -- every result includes meta.attempts detailing route, status, duration, and error codes per try
  • CLI and library API -- use ghx run <capability> from the terminal or executeTask() from code; createGithubClientFromToken() sets up a client in one line
  • Agent interface -- @ghx-dev/core/agent exports createExecuteTool, listCapabilities, explainCapability, and MAIN_SKILL_TEXT for agent frameworks
  • Capability discovery -- ghx capabilities list and ghx capabilities explain <id> for interactive exploration
  • Agent onboarding -- ghx setup --platform claude-code installs ghx as a project skill with a single command
  • GraphQL + CLI adapters -- dual execution paths with automatic fallback across routes
  • GitHub Enterprise support -- GH_HOST and GITHUB_GRAPHQL_URL for enterprise endpoints

  Quick Start

  npx @ghx-dev/core capabilities list
  npx @ghx-dev/core run repo.view --input '{"owner":"aryeko","name":"ghx"}'