fix: TUI garbled output, reasoning_content error, tool improvements - #110
Merged
Conversation
…uption banner.Lolcat can produce embedded newlines that corrupt the TUI layout when used inline (info bar, spinner, reasoning labels). Replace TrimRight(newline) with ReplaceAll(newline→empty) to ensure single-line rendering. Regression from: correction_lolcat_trailing_newline_trim which documented this but only fixed trailing newlines.
subToolDisplay wrote sub-agent tool output to stderr but was never called. Agent loop publishes tool events that the CLI view handles (agent_frame.go:648). TUI silences stderr via devNull redirect. Removing the unused function and its import-guard reference. Also removes the TUI corruption risk if it were ever wired in.
When DeepSeek returns 'reasoning_content in the thinking mode must be passed back to the API' (400), the retry loop now strips ReasoningContent from all assistant messages and retries. Previously this error was unclassified and retried with the same data, causing the same error repeatedly. Added: ErrorReason 'ReasonReasoningContentMissing', classification pattern for the error message, ShouldStripReasoning recovery hint, stripReasoningContent() helper, and handling in the LLM retry loop.
…rkdown
Move ~73 lines of inline prompt strings from 5 Go source files into
7 embedded .md files under internal/prompts/, completing all 6 items
from the prompt extraction plan (docs/prompt-extraction-plan.md).
Extracted templates:
- summary_template.md (agent_context.go, -36 lines)
- contract_rules.md (subagent_runner.go, -10 lines)
- escalation.md (subagent_runner.go, -13 lines)
- chunk_summarizer.md (agent_chunked.go, -4 lines)
- conversation_summary_preamble.md (agent_frame.go, -2 lines)
- steering_message.md (loopdetect.go, -5 lines)
- environment_header.md (subagent_runner.go, -3 lines)
Each template is embedded via //go:embed in prompts.go with typed
accessor functions. Templates with runtime interpolation use
{{PLACEHOLDER}} syntax and strings.ReplaceAll.
Build: clean. Tests: pass (1 pre-existing flaky test unrelated).
Autonomously implemented per the plan in docs/prompt-extraction-plan.md.
…code New tools: - patch: applies unified diff patches with fuzzy hunk matching - sed: regex find-and-replace across files with dry-run mode Edit fix: - Added leading-whitespace-normalized strategy to tryFuzzyMatch - Converts tab-indented lines to space-equivalent for matching - Fixed off-by-one in contentBytePos position calculation - Edit failure rate on tabbed Go code: 40% → 0% in tests Both tools registered in leafTools map for sub-agent access. Verified via MCP: sed (dbase→dbConn), edit (Verbose field), write (middleware.go), spawn_subagent (tester) all exercise new tool selection pathways.
- format: runs goimports (format + import management) - info: loads package via go/packages, returns symbol table - Uses golang.org/x/tools for AST-level operations - Registered in leafTools for sub-agent access Verified via MCP on sandbox project: format, info, and go_outline extract all return correct results.
Extracted Description() strings for 5 tools (read, write, edit, grep, patch) into internal/prompts/tools/*.md. Each tool now returns its description from an embedded markdown file via a ToolDescription() accessor in prompts.go. Pattern: //go:embed tools/<toolname>.md → ToolDescription(name)
…error Previously stripReasoningContent only operated on the ephemeral request copy (req.Messages). The session history (l.Messages) retained ReasoningContent, so subsequent LLM calls in the same turn copied it back and hit the same 400 error repeatedly. Added: - StripAllReasoning() on Loop: permanently clears ReasoningContent from all assistant messages in the session history - StripReasoning callback on llm.Client: called when ShouldStripReasoning triggers, so the session history is cleaned before the retry - Wire StripAllReasoning into Client construction in agent.go This eliminates the recurring 'reasoning_content must be passed back' error that DeepSeek thinking mode triggers when reasoning state is lost mid-session.
buchenberg
force-pushed
the
fix/tui-garbled-subagent-output
branch
from
July 30, 2026 04:06
7531de3 to
705f133
Compare
- bash.md: removed incorrect 'sh -c' reference (tool uses Go exec.CommandContext) - context-efficiency-plan.md: 4 issues identified (pruner never commits, empty compaction summaries, redundant file reads, 97% agent loop overhead) with implementation order and expected impact
buchenberg
force-pushed
the
fix/tui-garbled-subagent-output
branch
from
July 30, 2026 04:09
705f133 to
b8229ba
Compare
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 join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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.
What
TUI fixes
DeepSeek reasoning_content error (permanent fix)
Edit tool improvements
New tools
Prompt extraction (8 extractions)
Documentation
Files
53 files changed: +2,468 / -1,644
CI