agent-harness 0.29.0 → 0.31.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: e929940987f238b75b654eadf2d882b0a554fb755c3badac115c7ae218875677
-  data.tar.gz: a83fefa5de68968675304ed70316ccfa37a7b3063bd58a1edf3b8c493f017daf
+  metadata.gz: 1e4fa0ebff9c67559f7294ebe65550e75629b9c21d95f2e04155f944f9d4e5de
+  data.tar.gz: bed581986f9471514de4488702b8c120c830887835c393cb0c6ad2f5d42b974b
 SHA512:
-  metadata.gz: a08f5ce50f0b0c200034d53908ded075808f7a2e374aff7365d54240cf1dc291d1cde28d4e1d4e76cc2d16273aa30bb17f81384243d626ce60fd5c5fd6d2f702
-  data.tar.gz: 615fec572434e6a0653f3ae3cd1723f551d7578eb30ef2f1ee0a2714c6221be8acd29c06e3830d10f0e88c040c31faab51658c26b09b79e18215f295587c644e
+  metadata.gz: 480a0bffe3625ea4c7e43bdcb1cfb8857798156be26a69f23d88e38443e8b75682aa0125b5ca2af6e94e967839129482c1607779a01e7bf76d5bb673bebb8888
+  data.tar.gz: dc36eb56d5b62dd7956cf311cd0d9635dbab456f02550ea3cc14fdb304488e38aeda33a5a36234aae7fd31b3e47783fcf69698850131cb6e08ced1326e0b2d2f

data/.release-please-manifest.json

@@ -1,3 +1,3 @@
 {
-  ".": "0.29.0"
+  ".": "0.31.0"
 }

data/CHANGELOG.md

@@ -5,6 +5,27 @@
 * add runner model compatibility contract (`AgentHarness.model_compatibility`) with structured `ModelCompatibility::Result` outcomes. Codex exposes static facts for CLI-gated models (e.g. `gpt-5.5` requires Codex CLI `>= 0.116.0`), a baseline supported-model list, supported auth modes, and a `DEFAULT_COMPATIBLE_MODEL_ID` fallback so downstream orchestrators can validate tier/model assignments before scheduling agent runs ([#259](https://github.com/viamin/agent-harness/issues/259)).
 * **auth:** add provider-owned PKCE code-exchange API for Claude OAuth (`AgentHarness::Authentication.exchange_code`). Takes an authorization code plus PKCE verifier (and `redirect_uri`/`client_id`), posts an `authorization_code` grant to the Claude token endpoint, and persists the resulting access/refresh tokens in the native `claudeAiOauth` shape. Adds `exchange_code_supported?` and a `code_exchange` key to `auth_capabilities` ([#266](https://github.com/viamin/agent-harness/issues/266)).
 
+## [0.31.0](https://github.com/viamin/agent-harness/compare/agent-harness/v0.30.0...agent-harness/v0.31.0) (2026-07-16)
+
+
+### Features
+
+* release agent-harness 0.31.0 with omp support ([#302](https://github.com/viamin/agent-harness/issues/302)) ([7a82fbc](https://github.com/viamin/agent-harness/commit/7a82fbc5cc627eaa10c678e3e2b998fc0ce100f2))
+
+## [0.31.0](https://github.com/viamin/agent-harness/compare/agent-harness/v0.30.0...agent-harness/v0.31.0) (2026-07-16)
+
+
+### Features
+
+* release a single `agent-harness` gem version that consolidates the full `:omp` provider contract for downstream consumers: distinct `:omp` provider metadata separate from `:pi`, the install/runtime contract for `@oh-my-pi/pi-coding-agent` `17.0.1`, the Bun runtime floor `>= 1.3.14` (pinned install target `1.3.14`), the smoke-test contract, and regression coverage for the public `AgentHarness` APIs ([#297](https://github.com/viamin/agent-harness/issues/297))
+
+
+## [0.30.0](https://github.com/viamin/agent-harness/compare/agent-harness/v0.29.0...agent-harness/v0.30.0) (2026-07-16)
+
+
+### Features
+
+* Implement Oh My Pi execution, auth, and smoke-test contract ([#300](https://github.com/viamin/agent-harness/issues/300)) ([dc3f5ec](https://github.com/viamin/agent-harness/commit/dc3f5ecce4508f103a8b769727e31aaa3d694c7e))
 
 ## [0.29.0](https://github.com/viamin/agent-harness/compare/agent-harness/v0.28.6...agent-harness/v0.29.0) (2026-07-16)
 

data/lib/agent_harness/providers/omp.rb

@@ -31,6 +31,71 @@ module AgentHarness
       SUPPORTED_BUN_VERSION = "1.3.14"
       BUN_REQUIREMENT_STRING = ">= #{SUPPORTED_BUN_VERSION}".freeze
 
+      # Smoke-test contract suitable for container health probes.
+      #
+      # Oh My Pi runs non-interactively via `-p` with `--no-session`, so the
+      # probe is a single deterministic round trip. The contract is tuned for
+      # a tight probe budget: a short prompt that elicits a predictable reply
+      # and a bounded timeout that accommodates a cold Bun runtime start.
+      SMOKE_TEST_CONTRACT = {
+        prompt: "Reply with exactly OK.",
+        expected_output: "OK",
+        timeout: 30,
+        require_output: true,
+        success_message: "Oh My Pi smoke test passed"
+      }.freeze
+
+      # Oh My Pi is a multi-provider CLI: it routes to a backend through
+      # `--provider` and surfaces that backend's error vocabulary. The
+      # patterns below extend the shared HTTP-style set with Oh My Pi /
+      # upstream-Pi specific phrasing observed for auth expiry, quota and
+      # rate-limit surfaces, model resolution failures, and transient
+      # network faults.
+      ERROR_PATTERNS = {
+        rate_limited: COMMON_ERROR_PATTERNS[:rate_limited],
+        auth_expired: COMMON_ERROR_PATTERNS[:auth_expired] + [
+          /\b401\b/,
+          /session.*(?:expired|invalid)/i,
+          /log(?:ged)?.?in.*required/i,
+          /api.?key.*(?:invalid|missing|expired|revoked)/i,
+          /token.*(?:expired|invalid|revoked)/i,
+          /credentials.*(?:expired|invalid|missing)/i
+        ],
+        quota_exceeded: COMMON_ERROR_PATTERNS[:quota_exceeded],
+        # Model resolution is surfaced by both the omp CLI (unknown
+        # `--model`/`--provider` combination) and the upstream backend.
+        # Treat it as a distinct, non-retryable category so callers can
+        # fall back to a known-good model instead of retrying the same id.
+        model_not_found: [
+          /model.*not.*found/i,
+          /no.*such.*model/i,
+          /unknown.*model/i,
+          /model.*does.*not.*exist/i,
+          /invalid.*model/i,
+          /model.*unavailable/i,
+          /provider.*does.*not.*support.*model/i
+        ],
+        transient: COMMON_ERROR_PATTERNS[:transient] + [
+          /connection.*reset/i,
+          /econnreset/i,
+          /socket.*hang.*up/i,
+          /network.*error/i
+        ]
+      }.tap { |patterns| patterns.each_value(&:freeze) }.freeze
+
+      # Non-actionable output that Oh My Pi writes while the Bun runtime and
+      # the agent bootstrap. Downstream consumers use these to filter probe
+      # and run output so startup banners do not masquerade as failures.
+      NOISY_OUTPUT_PATTERNS = [
+        /oh.?my.?pi\b/i,
+        /pi(?:-coding.?agent)?\s+v?\d+\.\d+/i,
+        /\bbun\s+v?\d+\.\d+/i,
+        /\bloading/i,
+        /\binitializing/i,
+        /\bwarming.?up/i,
+        /fetching.*model/i
+      ].freeze
+
       class << self
         def provider_name
           :omp
@@ -49,7 +114,22 @@ module AgentHarness
           {
             auth: {
               service: :omp,
-              api_family: :multi_provider
+              api_family: :multi_provider,
+              # Oh My Pi routes to a backend through `--provider` and reads
+              # that backend's API key from its conventional env var. The
+              # harness never reuses the upstream Pi credential store
+              # (paid_pi_auth_entry); callers pass backend keys per request
+              # through ProviderRuntime#env, which build_env materializes into
+              # the subprocess environment. This keeps the omp runner an
+              # independent entity from :pi.
+              #
+              # Deliberately do NOT advertise a harness-managed credential
+              # store here. Authentication has no omp read/write/validate path,
+              # so auth_status(:omp) reports "not implemented"; exposing
+              # credential_store would imply a session store that does not
+              # exist and let callers infer harness-managed session auth.
+              # Surface the implemented per-request env model instead.
+              api_key_source: :provider_runtime_env
             }
           }
         end
@@ -158,7 +238,7 @@ module AgentHarness
         end
 
         def smoke_test_contract
-          Base::DEFAULT_SMOKE_TEST_CONTRACT
+          SMOKE_TEST_CONTRACT
         end
       end
 
@@ -202,13 +282,60 @@ module AgentHarness
           # Oh My Pi's non-interactive CLI currently exposes only text print mode.
           # Keep JSON mode disabled until the CLI ships a structured output flag.
           json_mode: false,
+          # MCP capability decision: the omp non-interactive print-mode path
+          # (`omp --no-session -p`) does not accept an MCP server config flag
+          # in the harness's supported CLI version. Keep MCP disabled here so
+          # callers do not attempt to attach servers that the runtime would
+          # reject. Revisit once omp ships a stable `--mcp-config` flag.
           mcp: false,
           dangerous_mode: false
         }
       end
 
       def error_patterns
-        COMMON_ERROR_PATTERNS
+        ERROR_PATTERNS
+      end
+
+      # Downstream-facing error classification. Augments the shared quota set
+      # with auth-expiry, model-resolution, and authentication-specific
+      # phrasing so consumers can route omp failures without re-deriving the
+      # CLI vocabulary. The inherited `:quota` set is deliberately preserved
+      # (not overridden) so omp's multi-provider backends surface their full
+      # credit/balance vocabulary (requires more credits, insufficient
+      # balance, spend limit reached, billing limit, etc.).
+      def error_classification_patterns
+        super.merge(
+          auth_expired: ERROR_PATTERNS[:auth_expired],
+          authentication: [
+            /api.?key.*not.*(?:set|configured)/i,
+            /no.*api.?key/i,
+            /missing.*credentials/i,
+            /log(?:ged)?.?in.*required/i
+          ],
+          model_not_found: ERROR_PATTERNS[:model_not_found]
+        )
+      end
+
+      # Oh My Pi emits a startup banner (version, Bun runtime, model fetch)
+      # that is not actionable. Expose it so callers can strip probe noise.
+      def noisy_error_patterns
+        NOISY_OUTPUT_PATTERNS
+      end
+
+      def translate_error(message)
+        if ERROR_PATTERNS[:model_not_found].any? { |p| message.match?(p) }
+          "Oh My Pi could not resolve the requested model. Check --model/--provider."
+        elsif /api.?key.*not.*(?:set|configured)/i.match?(message) || /no.*api.?key/i.match?(message)
+          "Oh My Pi API key not set for the selected provider."
+        else
+          message
+        end
+      end
+
+      # Oh My Pi runs stateless through `--no-session`; the harness does not
+      # expose session persistence, so callers cannot resume a session.
+      def supports_sessions?
+        false
       end
 
       def supports_tool_control?
@@ -219,6 +346,36 @@ module AgentHarness
         :oauth
       end
 
+      # Conventional backend API-key env vars the omp CLI reads. Oh My Pi is
+      # multi-provider: the effective var is the one matching the selected
+      # `--provider`, supplied per request through ProviderRuntime#env.
+      def api_key_env_var_names
+        [
+          "ANTHROPIC_API_KEY",
+          "OPENAI_API_KEY",
+          "GEMINI_API_KEY",
+          "GOOGLE_API_KEY",
+          "XAI_API_KEY",
+          "DEEPSEEK_API_KEY",
+          "OPENROUTER_API_KEY",
+          "GROQ_API_KEY",
+          "MISTRAL_API_KEY"
+        ]
+      end
+
+      # omp routes through `--provider` rather than an env-driven proxy/base
+      # URL, so there are no known proxy header vars to scrub when a caller
+      # supplies its own key.
+      def api_key_unset_vars
+        []
+      end
+
+      # When running against an OAuth/subscription session, drop backend
+      # API-key env vars so the CLI prefers the stored session credentials.
+      def subscription_unset_vars
+        api_key_env_var_names
+      end
+
       def execution_semantics
         {
           prompt_delivery: :flag,

data/lib/agent_harness/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module AgentHarness
-  VERSION = "0.29.0"
+  VERSION = "0.31.0"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: agent-harness
 version: !ruby/object:Gem::Version
-  version: 0.29.0
+  version: 0.31.0
 platform: ruby
 authors:
 - Bart Agapinan