mirror of
https://github.com/NousResearch/hermes-agent.git
synced 2026-05-18 04:41:56 +00:00
fix(copilot-acp): tighten deprecation detection + sharpen GitHub Models 413 hint
Follow-up improvements on top of @konsisumer's cherry-picked fix for #10648: 1. Deprecation patterns required BOTH a product fingerprint ('gh-copilot') and a deprecation marker. The previous list included 'copilot-cli' and bare 'deprecation', which would false-positive on stderr from the NEW @github/copilot CLI — whose repo is literally github.com/github/copilot-cli and which legitimately surfaces those substrings in its own messages. 2. Replace the deprecation hint. The user in #10648 installed 'gh extension install github/gh-copilot' (the deprecated extension) thinking that's what ACP mode uses, when ACP actually spawns the new 'copilot' binary from '@github/copilot'. The hint now points users at the correct install command ('npm install -g @github/copilot') with the new CLI's repo URL, and demotes provider-switching to a fallback alternative. 3. Change _URL_TO_PROVIDER value for models.inference.ai.azure.com from the 'github-models' alias to the canonical 'copilot' provider id, matching the convention used by every other entry in the table. 4. Sharpen the 413 hint message. The free tier's ~8K cap is below the system-prompt floor, so this endpoint is fundamentally incompatible with an agentic loop — not a 'use a different URL' problem. Tests: - New parametrized false-positive coverage for the new CLI's stderr shape. - Updated assertion to require canonical 'copilot' provider mapping. - All 14 deprecation/URL tests pass.
This commit is contained in:
teknium1
Teknium
parent
b85b938b1f
commit
374dc81c23
4 changed files with 84 additions and 38 deletions
|
|
@ -30,15 +30,29 @@ _DEFAULT_TIMEOUT_SECONDS = 900.0
|
|||
_TOOL_CALL_BLOCK_RE = re.compile(r"<tool_call>\s*(\{.*?\})\s*</tool_call>", re.DOTALL)
|
||||
_TOOL_CALL_JSON_RE = re.compile(r"\{\s*\"id\"\s*:\s*\"[^\"]+\"\s*,\s*\"type\"\s*:\s*\"function\"\s*,\s*\"function\"\s*:\s*\{.*?\}\s*\}", re.DOTALL)
|
||||
|
||||
# Patterns that indicate the gh-copilot CLI extension has been deprecated.
|
||||
_DEPRECATION_PATTERNS = (
|
||||
# Stderr fingerprint of the deprecated `gh copilot` CLI extension
|
||||
# (https://github.blog/changelog/2025-09-25-upcoming-deprecation-of-gh-copilot-cli-extension).
|
||||
# We require BOTH the literal product name ("gh-copilot") AND a deprecation
|
||||
# marker, so generic stderr from the NEW `@github/copilot` CLI — whose repo
|
||||
# is github.com/github/copilot-cli and which legitimately mentions "copilot-cli"
|
||||
# in its own banners and error messages — doesn't get misclassified as the
|
||||
# deprecated extension.
|
||||
_DEPRECATION_REQUIRED = ("gh-copilot",)
|
||||
_DEPRECATION_MARKERS = (
|
||||
"has been deprecated",
|
||||
"no commands will be executed",
|
||||
"deprecation",
|
||||
"copilot-cli",
|
||||
)
|
||||
|
||||
|
||||
def _is_gh_copilot_deprecation_message(stderr_text: str) -> bool:
|
||||
"""True iff stderr looks like the deprecated gh-copilot extension's banner."""
|
||||
|
||||
lower = stderr_text.lower()
|
||||
if not any(req in lower for req in _DEPRECATION_REQUIRED):
|
||||
return False
|
||||
return any(marker in lower for marker in _DEPRECATION_MARKERS)
|
||||
|
||||
|
||||
def _resolve_command() -> str:
|
||||
return (
|
||||
os.getenv("HERMES_COPILOT_ACP_COMMAND", "").strip()
|
||||
|
|
@ -514,16 +528,19 @@ class CopilotACPClient:
|
|||
|
||||
stderr_text = "\n".join(stderr_tail).strip()
|
||||
if proc.poll() is not None and stderr_text:
|
||||
stderr_lower = stderr_text.lower()
|
||||
if any(pat in stderr_lower for pat in _DEPRECATION_PATTERNS):
|
||||
if _is_gh_copilot_deprecation_message(stderr_text):
|
||||
raise RuntimeError(
|
||||
"The gh-copilot CLI extension has been deprecated by GitHub and "
|
||||
"can no longer be used for ACP mode.\n\n"
|
||||
"Alternatives:\n"
|
||||
" 1. Use the GitHub Copilot provider instead of ACP mode:\n"
|
||||
" hermes setup → select 'GitHub Copilot' (uses Copilot Chat API)\n"
|
||||
" 2. Set HERMES_COPILOT_ACP_COMMAND to point to a compatible ACP server\n"
|
||||
" 3. Use a different provider (e.g. OpenAI, Anthropic, Nous)\n\n"
|
||||
"Hermes ACP mode requires the NEW GitHub Copilot CLI "
|
||||
"(github.com/github/copilot-cli), but the binary it just "
|
||||
"spawned is the deprecated `gh copilot` extension.\n\n"
|
||||
"Install the new CLI:\n"
|
||||
" npm install -g @github/copilot\n"
|
||||
" # then verify with: copilot --help\n\n"
|
||||
"If `copilot` already resolves to the new CLI but you still see this,\n"
|
||||
"point Hermes at it explicitly:\n"
|
||||
" export HERMES_COPILOT_ACP_COMMAND=/path/to/new/copilot\n\n"
|
||||
"Alternative: use the `copilot` provider (no ACP, hits the Copilot API\n"
|
||||
"directly with a Copilot subscription token) via `hermes setup`.\n\n"
|
||||
f"Original error:\n{stderr_text}"
|
||||
)
|
||||
raise RuntimeError(f"Copilot ACP process exited early: {stderr_text}")
|
||||
|
|
|
|||
|
|
@ -358,7 +358,12 @@ _URL_TO_PROVIDER: Dict[str, str] = {
|
|||
"api.deepseek.com": "deepseek",
|
||||
"api.githubcopilot.com": "copilot",
|
||||
"models.github.ai": "copilot",
|
||||
"models.inference.ai.azure.com": "github-models",
|
||||
# GitHub Models free tier (Azure-hosted prototyping endpoint) — same
|
||||
# canonical provider as the Copilot API. Hard per-request token cap
|
||||
# (often 8K) makes it unusable for Hermes' system prompt, but mapping
|
||||
# it here lets us recognize the endpoint and emit a targeted hint
|
||||
# instead of falling through the unknown-custom-endpoint path.
|
||||
"models.inference.ai.azure.com": "copilot",
|
||||
"api.fireworks.ai": "fireworks",
|
||||
"opencode.ai": "opencode-go",
|
||||
"api.x.ai": "xai",
|
||||
|
|
|
|||
22
run_agent.py
22
run_agent.py
|
|
@ -14185,29 +14185,35 @@ class AIAgent:
|
|||
}
|
||||
|
||||
# Actionable hint for GitHub Models (Azure) 413 errors.
|
||||
# The free tier enforces a hard 8K token limit per request,
|
||||
# which Hermes' system prompt alone can exceed. Compression
|
||||
# won't help — surface a clear message so the user doesn't
|
||||
# wait through three futile compression attempts.
|
||||
# The free tier enforces a hard 8K token cap per request,
|
||||
# which Hermes' system prompt + tool schemas alone exceed.
|
||||
# Compression can't help — the floor is the system prompt
|
||||
# itself, not the conversation — so surface a clear "not
|
||||
# compatible" message instead of looping into three futile
|
||||
# compression attempts.
|
||||
if (
|
||||
status_code == 413
|
||||
and isinstance(_base, str)
|
||||
and "models.inference.ai.azure.com" in _base
|
||||
):
|
||||
self._vprint(
|
||||
f"{self.log_prefix} 💡 GitHub Models (Azure) enforces a hard per-request token limit (often 8K).",
|
||||
f"{self.log_prefix} 💡 GitHub Models free tier (models.inference.ai.azure.com) caps every",
|
||||
force=True,
|
||||
)
|
||||
self._vprint(
|
||||
f"{self.log_prefix} Hermes' system prompt alone may exceed this limit. This endpoint is not",
|
||||
f"{self.log_prefix} request at ~8K tokens. Hermes' system prompt + tool schemas baseline",
|
||||
force=True,
|
||||
)
|
||||
self._vprint(
|
||||
f"{self.log_prefix} compatible with Hermes Agent. Use https://models.github.ai or the GitHub",
|
||||
f"{self.log_prefix} exceeds that floor, so this endpoint cannot run an agentic loop.",
|
||||
force=True,
|
||||
)
|
||||
self._vprint(
|
||||
f"{self.log_prefix} Copilot provider instead, which have higher token limits.",
|
||||
f"{self.log_prefix} Use the `copilot` provider with a Copilot subscription token (`hermes",
|
||||
force=True,
|
||||
)
|
||||
self._vprint(
|
||||
f"{self.log_prefix} setup` → GitHub Copilot), or pick any other provider.",
|
||||
force=True,
|
||||
)
|
||||
|
||||
|
|
|
|||
|
|
@ -2,11 +2,12 @@
|
|||
|
||||
import pytest
|
||||
|
||||
from agent.copilot_acp_client import _DEPRECATION_PATTERNS
|
||||
from agent.copilot_acp_client import _is_gh_copilot_deprecation_message
|
||||
|
||||
|
||||
class TestDeprecationPatternDetection:
|
||||
"""Verify that stderr messages from a deprecated gh-copilot CLI are caught."""
|
||||
"""Verify that stderr from the deprecated `gh copilot` extension is caught
|
||||
without false-positiving on the new `@github/copilot` CLI."""
|
||||
|
||||
_REAL_DEPRECATION_STDERR = (
|
||||
"The gh-copilot extension has been deprecated in favor of the newer "
|
||||
|
|
@ -18,25 +19,40 @@ class TestDeprecationPatternDetection:
|
|||
)
|
||||
|
||||
def test_real_deprecation_message_matches(self):
|
||||
lower = self._REAL_DEPRECATION_STDERR.lower()
|
||||
assert any(pat in lower for pat in _DEPRECATION_PATTERNS)
|
||||
assert _is_gh_copilot_deprecation_message(self._REAL_DEPRECATION_STDERR)
|
||||
|
||||
@pytest.mark.parametrize(
|
||||
"stderr_line",
|
||||
"stderr_text",
|
||||
[
|
||||
"The gh-copilot extension has been deprecated",
|
||||
"No commands will be executed.",
|
||||
"See deprecation notice at ...",
|
||||
"Install copilot-cli instead",
|
||||
# The deprecation banner uses both halves of the fingerprint.
|
||||
"The gh-copilot extension has been deprecated.",
|
||||
"gh-copilot: no commands will be executed.",
|
||||
# Mixed casing — match is case-insensitive.
|
||||
"The GH-Copilot Extension HAS BEEN DEPRECATED.",
|
||||
],
|
||||
)
|
||||
def test_individual_patterns_match(self, stderr_line: str):
|
||||
lower = stderr_line.lower()
|
||||
assert any(pat in lower for pat in _DEPRECATION_PATTERNS)
|
||||
def test_genuine_deprecation_variants_match(self, stderr_text: str):
|
||||
assert _is_gh_copilot_deprecation_message(stderr_text)
|
||||
|
||||
def test_normal_stderr_does_not_match(self):
|
||||
normal = "Error: connection refused"
|
||||
assert not any(pat in normal.lower() for pat in _DEPRECATION_PATTERNS)
|
||||
@pytest.mark.parametrize(
|
||||
"stderr_text",
|
||||
[
|
||||
# Generic errors — no fingerprint at all.
|
||||
"Error: connection refused",
|
||||
"",
|
||||
# The NEW @github/copilot CLI's repo is github.com/github/copilot-cli.
|
||||
# Its stderr can legitimately mention "copilot-cli" or "deprecation"
|
||||
# in unrelated contexts; neither alone should trip the detector.
|
||||
"copilot-cli: failed to authenticate with the API",
|
||||
"warning: the --foo flag is scheduled for deprecation in v3",
|
||||
"See https://github.com/github/copilot-cli/issues for support",
|
||||
# Half the fingerprint without the other half.
|
||||
"gh-copilot: command not found",
|
||||
"extension has been deprecated (some other extension)",
|
||||
],
|
||||
)
|
||||
def test_does_not_false_positive(self, stderr_text: str):
|
||||
assert not _is_gh_copilot_deprecation_message(stderr_text)
|
||||
|
||||
|
||||
class TestGitHubModelsAzureUrl:
|
||||
|
|
@ -45,7 +61,9 @@ class TestGitHubModelsAzureUrl:
|
|||
def test_url_to_provider_contains_azure_models(self):
|
||||
from agent.model_metadata import _URL_TO_PROVIDER
|
||||
|
||||
assert _URL_TO_PROVIDER.get("models.inference.ai.azure.com") == "github-models"
|
||||
# Maps to the canonical "copilot" provider (same convention as the
|
||||
# other GitHub-family entries) — not the "github-models" alias.
|
||||
assert _URL_TO_PROVIDER.get("models.inference.ai.azure.com") == "copilot"
|
||||
|
||||
def test_is_github_models_base_url_recognises_azure(self):
|
||||
from hermes_cli.models import _is_github_models_base_url
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue