Commit Graph

52 Commits

Author SHA1 Message Date
CJACK
77a47ada4e Fix tool detection when unclosed backtick precedes tool call
Handles cases where a stray backtick opens an inline code span but is never closed.
Previously, any subsequent XML tool tag was treated as inside markdown code and ignored.
Now, tool tags are detected after an unclosed backtick, and the markdown state is reset
when the backtick is confirmed to be literal text at stream boundaries.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-05-10 18:41:51 +08:00
CJACK
e393110121 fix toolcall inline code and query redaction 2026-05-10 18:02:54 +08:00
CJACK
cee8757d14 revert: replace fullwidth pipe | with halfwidth | in DSML tool markup
PR #460 introduced fullwidth pipe characters (|) in DSML tool call formatting
to improve parsing robustness, but models exposed to these fullwidth pipes in
system prompts exhibit significantly higher rates of tool output hallucinations.
Reverting to halfwidth pipes (|) drastically reduces tokenizer/perplexity-driven
hallucinations while retaining the existing confusable-hardening in the parser.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-05-10 15:18:54 +08:00
Your Name
196e3c46f6 feat(toolcall): harden confusable candidate spans 2026-05-10 09:27:30 +07:00
CJACK
1aa791ec3a feat: support PascalCase local-name drift in DSML tool markup parsing
Detect camelCase→PascalCase boundaries between arbitrary prefixes and fixed
local names (tool_calls/invoke/parameter), so that fused forms like
<DSmartToolCalls> are recognized without explicit separator characters.
Also add the underscore-free alias "toolcalls" as a valid DSML local name.
Includes lookalike rejection tests to ensure near-matches like
<DSmartToolCallsExtra> are not falsely accepted.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-05-10 04:52:19 +08:00
CJACK
247fc7c788 refactor: unify tool markup pipe and CDATA separator into general-purpose separator detector
Replace the hardcoded isToolMarkupPipe (matching |, |, ␂, \x02, !) and
isToolCDATAOpenSeparator (exclusion-based) with a single isToolMarkupSeparator
that treats any Unicode punctuation outside structural characters as a valid
DSML separator. This eliminates the need for a per-character allowlist — novel
separators like ※ are automatically supported without code changes. Also
removes the unused cdataPattern regexp and updates docs to use "non-structural
separator" terminology.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-05-10 04:24:10 +08:00
CJACK
7a28b9e265 feat: improve CDATA and DSML tag parsing robustness with support for fullwidth-bang, ideographic-comma, and extended quote/separator normalization. 2026-05-10 03:41:55 +08:00
CJACK
61d42f8b72 feat: add support for CJK angle bracket and trailing attribute separator drift in DSML tool parsing 2026-05-10 01:54:31 +08:00
CJACK
740a78ad5a refactor: allow and preserve empty tool parameter values while updating sieve to release malformed XML as text 2026-05-10 01:05:18 +08:00
CJACK
3cc7f469f3 feat: implement support for arbitrary tool markup prefixes and control character separators in tool sieve parsing 2026-05-10 00:19:03 +08:00
CJACK
7c66742a19 refactor: unify empty-output retry logic into shared completionruntime package and normalize protocol adapter boundary. 2026-05-10 00:10:53 +08:00
CJACK
067cf465bb feat: integrate reasoning content into assistant tool-call messages and improve tool markup parsing for prompt compatibility 2026-05-09 23:16:07 +08:00
CJACK
2f7cb473fc feat: support hyphenated DSML tag variants in tool-call parsing
Add compatibility for <dsml-tool-calls>/<dsml-invoke>/<dsml-parameter>
tag forms alongside the canonical pipe-prefixed DSML shell. Hyphenated
forms only activate when a DSML prefix is detected, preventing false
matches on bare XML lookalikes. Go and Node parsers aligned, with tests
covering here-doc CDATA, streaming sieve, and negative lookalike cases.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-05-03 03:09:10 +08:00
CJACK
1c38709d32 feat: add support for parsing loose JSON lists into arrays in tool call parameters 2026-05-02 03:26:43 +08:00
CJACK
fd0ec29991 refactor: generalize DSML tag parsing to tolerate model noise; split tiktoken by build tags
Replace hardcoded DSML typo variant lists in Go/Node tool call parsers with
generalized prefix consumption that tolerates repeated leading <, repeated DSML
prefix noise, and trailing pipe terminators. Split tiktoken-dependent token
counting into a build-tagged file for non-cgo platform compatibility. Add /data
directory to Dockerfile for bind-mount permissions.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-05-01 15:17:11 +08:00
shern-point
f1926a6ced fix: normalize Vercel stream tool arguments by schema 2026-04-29 02:00:01 +08:00
CJACK
63271aea8c refactor: update tool call parsing and stream tool sieve logic
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-04-28 01:39:32 +08:00
CJACK
1e9170e385 补丁 2026-04-27 22:57:48 +08:00
CJACK
2d5d211a7a refactor: unify Go/Node XML tool markup scanning and expand DSML alias support
- Add shared ToolMarkupTag scanner (toolcalls_scan.go) replacing hardcoded alias tables
- Support DSML collapsed tag names (<DSMLtool_calls>, <DSMLinvoke>, <DSMLparameter>)
- Parse JSON literal values from parameter bodies (123→number, true→bool, null)
- Recover unclosed CDATA in final parse/flush via SanitizeLooseCDATA
- Align Go and Node implementations (scanToolMarkupTagAt, findMatchingToolMarkupClose)
- Reject bare <invoke> as unsupported syntax, only tool_calls wrapper triggers tool path
- Update API.md and toolcall-semantics.md documentation

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-04-27 17:53:59 +08:00
CJACK
70467054c3 fix: preserve partial-update fields for current_input_file and thinking_injection, expand DSML space-separator aliases
- Guard current_input_file.enabled / thinking_injection.{enabled,prompt} with hasNestedSettingsKey so partial updates don't overwrite omitted fields
- Expand DSML alias support to tolerate space-separated tags (e.g. <|dsml invoke>) alongside pipe-separated forms
- Sync Go sieve, Node sieve, toolcall parser, and tests for all new DSML variants
- Update API.md and toolcall-semantics.md with expanded alias coverage

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-04-27 15:06:44 +08:00
CJACK
a13293e113 feat: expand DSML tool-call alias and fence handling
Add support for DSML wrapper aliases (<dsml|tool_calls>, <|tool_calls>,
<|tool_calls>) alongside canonical XML. Normalize mixed DSML/canonical
tags instead of rejecting them. Add tilde fence (~~~) support, fix
nested fence and unclosed fence handling, support CDATA-protected fence
content, and skip prose mentions when scanning for real tool blocks.
Mirror all changes between Go and Node.js runtimes.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-04-27 13:39:50 +08:00
CJACK
40d5e3ebb5 测试DSML 2026-04-27 00:21:26 +08:00
CJACK
0fb1bc6611 工具优化 2026-04-26 09:44:59 +08:00
CJACK
7475defeca fix: align tool call protocol and thinking controls 2026-04-26 04:26:51 +08:00
CJACK
1b0e8cbadb Tighten XML tool call parsing and upstream empty handling 2026-04-26 01:17:16 +08:00
CJACK.
c291d333c4 feat: extract and inject assistant reasoning content into history split prompts 2026-04-22 19:56:28 +00:00
CJACK
f313d0068f feat: implement code fence awareness in tool sieve to prevent false-positive XML tool detection inside code blocks and refine prompt instructions. 2026-04-20 00:13:14 +08:00
CJACK
c945f49fc4 refactor: remove JSON-based tool call parsing from sieve and delete associated compatibility tests 2026-04-19 13:39:47 +08:00
CJACK.
bfca84c2c7 Align tool-call parsing across Go/JS and pass quality gates 2026-04-01 01:24:55 +08:00
CJACK.
1fe1240240 fix(js): prevent XML wrapper attribute tool_calls scan loop 2026-03-30 15:59:34 +08:00
CJACK.
775bf3b578 refactor(js): align tool-sieve segment start and tail window with go 2026-03-30 15:41:26 +08:00
CJACK.
ab3943ebeb test(js): cover numbered planning prose around tool calls 2026-03-30 15:39:09 +08:00
CJACK.
6efba7b2e4 fix(js): avoid false tool-call capture on plain tool_calls prose 2026-03-30 12:51:33 +08:00
CJACK
30a53b6c43 refactor: remove legacy TOOL_CALL_HISTORY/TOOL_RESULT_HISTORY markers and consolidate tool call formatting into a new prompt package 2026-03-30 00:20:38 +08:00
CJACK
c3c644ff8c 111 2026-03-29 19:49:52 +08:00
CJACK.
97a81c4191 Harden toolcall leak interception for function-style payloads 2026-03-22 20:07:12 +08:00
CJACK.
b108a7915a Support nested fenced blocks in stream fence tracking 2026-03-22 15:12:55 +08:00
CJACK.
d530d25793 Expand history-sanitize boundary coverage for stream chunks 2026-03-22 00:57:13 +08:00
CJACK.
648bb74587 Fix streaming whitespace trim and capture TOOL_RESULT_HISTORY 2026-03-22 00:44:44 +08:00
CJACK.
6758514c61 chore: remove obsolete openai tool-history normalization helpers 2026-03-22 00:28:32 +08:00
CJACK.
7061094964 Fix fence-strip regression for closed code blocks before tool JSON 2026-03-21 17:39:08 +08:00
CJACK.
b8ff678f24 Align Vercel JS toolcall filtering with Go semantics 2026-03-21 00:23:22 +08:00
CJACK.
184a3d1e4e Sync Node tool-call parsing with aggressive fenced/mixed policy 2026-03-20 02:16:37 +08:00
CJACK.
145501d4a5 fix(tool-sieve): allow mixed prose + tool json interception 2026-03-20 01:15:32 +08:00
CJACK.
12d5f136d5 fix(toolcall): pass gates and align go/js multi-layer parser 2026-03-09 19:16:28 +08:00
CJACK.
60e9d707d4 Merge origin/dev into PR branch and resolve toolcall test conflicts 2026-03-08 01:10:53 +08:00
CJACK.
9b93badb57 Harden markup tag parsing to avoid mismatched-tag false positives 2026-03-08 00:55:32 +08:00
CJACK
93e9fb531d js对齐 2026-03-01 07:15:35 +08:00
CJACK
a302fb3c25 修复 2026-03-01 05:55:46 +08:00
CJACK
0d3d535c08 feat: prevent raw tool call JSON leakage for unknown or rejected tool calls and consolidate container publishing to GHCR. 2026-02-23 00:27:46 +08:00