Repository navigation
[pull] main from can1357:main - #104
Merged
Merged
Conversation
The user-scope guard read scope.project === false on an object that only
carries { cwd, user } or { cwd, project }, which works by accident via
undefined comparison. Name the scopes explicitly so the project-capability
precedence check cannot silently invert.
Teach the native, .agent/.agents, .claude, .codex, and .gemini providers to load SYSTEM_TEMPLATE.md beside SYSTEM.md with a kind tag, and resolve the override from one loadCapability pass: project beats user, template beats literal within a scope. Ancestor walk-up now applies to templates, so a repo-root template wins from a nested cwd with no findConfigFile back-check. Discovered templates degrade instead of bricking startup: empty falls through to the literal (or bundled prompt), malformed warns and renders the bundled prompt. Explicit --system-prompt-template and SDK sources stay strict. xdev catalog names follow the rendered block 0: claimed only when it contains an xd:// reference.
SYSTEM.md is the long-established override; an existing literal keeps working until its author deliberately removes it in favor of a template. Precedence is now project literal, project template, user literal, user template. A same-scope malformed template beside a literal is never rendered, so the literal-malformed test asserts shadowing directly and the bundled-fallback test uses a lone template.
…mplate feat(coding-agent): support Handlebars system prompt overrides
The unified discovery commit pinned llama.cpp Qwen to qwen-chat-template (kwargs-only), but llama-server maps top-level OpenAI fields into the template natively: the twin emission is the documented contract. Restore thinking-format qwen + qwen-enable-thinking-false so enable_thinking, preserve_thinking, and 3.8 reasoning_effort ride both spellings. Update the discovery pin and the two stale discovered-model tests: the reviewed thinking-upgrade-neutral policy repairs spec.reasoning=false into a reasoning-capable model, so default requests take the implicit-disable path (explicit enable_thinking=false) rather than leaving the toggle unset.
fix(catalog): restore qwen top-level dialect for llama.cpp
- parsed Anthropic policy error codes from structured response bodies - classified organization OAuth denials as account-scoped policy errors - cleared sticky credentials before rotating to an eligible sibling
- narrowed overflow helpers to the evidence they actually consume - handled provider failures received before usage metadata exists
- added the fixture-driven native parser and renderer\n- switched utils and the class-based tui to native bindings\n- removed vendored typescript sources, fixtures, and notices
- classified chat, image, speech, search, judge, and local models - added TypeSafe and OpenRouter discovery with synthetic catalogs - compiled provider compatibility policy for the expanded catalog
- moved the no-auth sentinel into the shared retry module - reused catalog TypeSafe discovery parsing for judgment requests
- added model-kind and role-section tabs to the class model hub - separated chat and specialist roles in the legacy roles view - filtered role assignment candidates by catalog capability
- added image, web, speech, dictation, judge, memory, and local model routing - migrated legacy provider ordering settings into role fallback chains - added OpenRouter search and TypeSafe judgment discovery - validated kind-role fallbacks against credential-independent catalog slices
- Split provider helpers into focused import boundaries.\n- Converted built-in dispatch and consumers to static imports.
- Embedded large JSON assets through JSON.parse in release builds.\n- Kept lightweight helpers off heavy provider graphs and improved timing output.
- Dropped terminal priorities and lookaheads from sloppy.lark; Codex constrained decoding rejected them with "priority not supported yet". - Header terminals are now excluded from BODY_TEXT/AFTER_TEXT via terminal intersection and complement, which LLGuidance supports.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
See Commits and Changes for more details.
Created by
pull[bot] (v2.0.0-alpha.4)
Can you help keep this open source service alive? 💖 Please sponsor : )