xonecas.io
64875
Mc

mini-coder / commit 59cfedd

chore: clean up TODO.md — remove completed items, reduce verbosity

xonecas2026-03-20 59cfedd999379fb2f4275a156da816e1dba360ca parent a71484a

1 file changed

TODO.md+17−16
@@ -1,22 +1,23 @@
11 # TODO
22
3-- ~~mini coders uses 10x more tokens than claude using the anthropic subscription; this is likely an issue with caching not working correctly or our pruning resetting the cache.~~ Fixed: step pruning now anchors the tool-call window to the initial boundary (no prefix modification), preserves reasoning between steps, stabilizes cache breakpoints at last-user-message, and strips stale breakpoints. Falls back to full pruning past 200 messages.
3+## Bugs
4+
45 - CTRL+d hangs sometimes and doesn't exit.
6+- `supportsThinking()` trusts models.dev flag too broadly — sends adaptive thinking to models that reject it (e.g. `claude-haiku-4-5` via Zen returns 400).
7+- AI SDK expands `claude-3-5-haiku` → dated variant that Zen doesn't serve (404).
8+- Shell tool max-timeout leaves the parent terminal in a broken state.
9+- MiniMax sends empty text deltas in one-shot — renders as blank lines.
10+- Double prompt glyph after ESC — `turn-complete` newline + input loop prompt collide.
11+- Truncated tool commands cut mid-argument (e.g. `--new '{…`). Truncate at argument boundaries instead.
12+
13+## Features
514
6----
15+- Conversation summary on max context instead of just an error with `/new` suggestion.
16+- Derive `maxOutputTokens` from model info instead of hardcoding 16384.
17+- `/_debug` hidden command — snapshot recent logs/db into a report in cwd.
718
8-# Deferred fixes
19+## Refactors
920
10-- **Thinking effort breaks `claude-haiku-4-5` via Zen** — The model info cache (sourced from models.dev) marks `claude-haiku-4-5` as `reasoning: 1`. When a thinking effort is persisted (e.g. `medium`), the app sends `"thinking": {"type": "adaptive"}` which Zen (and likely the Anthropic API) rejects with a 400: _"adaptive thinking is not supported on this model"_. The `supportsThinking()` check is too broad — it trusts the models.dev flag but doesn't validate whether the specific model actual …
11-- **`zen/claude-3-5-haiku` 404** — The AI SDK auto-resolves `claude-3-5-haiku` to `claude-3-5-haiku-20241022`, which Zen doesn't serve. The Zen endpoint expects exactly `claude-3-5-haiku`. This is an SDK behaviour issue that may need a workaround or documentation.
12-- Several truncation helpers across the cli module files, consolidate in a cli/truncate.ts for rendering truncation.
13-- Consolidate `parseSkillFrontmatter` in `skills.ts` with `parseFrontmatter` in `frontmatter.ts` — they do the same YAML-lite parsing with different field sets.
14-- Status bar: combine model name with thinking effort into a single field (e.g. `provider/model medium`), and remove `showReasoning` from the status bar since it is already shown in the banner.
15-- When a shell tool call hits max timeout, the turn ends and the parent terminal is left in a broken state.
16-- **MiniMax empty lines in one-shot** — Investigate empty text deltas. May need to skip rendering deltas that are only whitespace at stream start.
17-- **Double prompt glyph after ESC** — The `turn-complete` handler writes a newline when `!renderedVisibleOutput`, then the input loop also writes its prompt. May need coordination to avoid the double prompt.
18-- **No conversation summary on max context** — The idea says "Stops with a conversation summary if max context is reached." Currently shows an error and suggests `/new` without generating a summary of the conversation.
19-- **Truncated tool commands** — Long `mc-edit` invocations truncate with `…` mid-argument (e.g. `--new '{…`). Could truncate at argument boundaries for clarity.
20-- `maxOutputTokens` is hardcoded at 16384 in `turn-request.ts`. Some models support higher output (e.g. Claude Opus 32k). Consider deriving from model info or making it configurable.
21-- `/_debug` hidden command that snapshots recent logs/db and creates a report in the cwd. For dev mostly but available to all. Do not list it anywhere, only documented here and in the code itself.
22-- Check this skill to learn about tmux and how we can use it to run manual tests. https://raw.githubusercontent.com/obra/superpowers-lab/refs/heads/main/skills/using-tmux-for-interactive-commands/SKILL.md
21+- Consolidate truncation helpers across cli modules into `cli/truncate.ts`.
22+- Consolidate `parseSkillFrontmatter` (skills.ts) with `parseFrontmatter` (frontmatter.ts).
23+- Status bar: merge model + thinking effort into one field, drop `showReasoning` (already in banner).