Uh oh!
There was an error while loading. Please reload this page.
docs: add clearer bedrock instructions - #40
Merged
Conversation
adamdotdevin
commented
May 20, 2025
Member
thanks a ton! |
randomm referenced
this pull request
in randomm/opencode
Jan 9, 2026
* fix(#41): correct Perplexity tool names across configs and prompts Replace incorrect tool references with actual MCP server tool names: - mcp_perplexity_research → perplexity-ask_deep_research - mcp_perplexity_ask → perplexity-ask_search, perplexity-ask_reason, perplexity-ask_deep_research Files updated: - opencode.work.json: 28 tool references across 14 agents - opencode.personal.json: 28 tool references across 14 agents - 18 prompt files with corrected Perplexity tool names This fixes tool invocation errors when agents attempt to use Perplexity research capabilities. * feat(#40): add research-backed boundary definitions to agent prompts Implement minimal, evidence-based prompt improvements to prevent agents from attempting GitHub operations outside their scope: PM prompt (4 lines): - Explicit coordination-only boundary - GitHub ops delegation to git-agent - Issue verification requirement Specialist prompts (6 lines each × 13): - Standardized scope boundary section - Do-not checklist (keyword format) - Cross-domain escalation protocol - Pre-work validation checklist Research basis: Perplexity research on prompting best practices found checklist/keyword format achieves 2-3x better compliance than narrative prose with fewer tokens. Files: 14 prompts updated, 97 lines total
randomm referenced
this pull request
in randomm/opencode
Jan 9, 2026
* fix(#41): correct Perplexity tool names across configs and prompts Replace incorrect tool references with actual MCP server tool names: - mcp_perplexity_research → perplexity-ask_deep_research - mcp_perplexity_ask → perplexity-ask_search, perplexity-ask_reason, perplexity-ask_deep_research Files updated: - opencode.work.json: 28 tool references across 14 agents - opencode.personal.json: 28 tool references across 14 agents - 18 prompt files with corrected Perplexity tool names This fixes tool invocation errors when agents attempt to use Perplexity research capabilities. * feat(#40): add research-backed boundary definitions to agent prompts Implement minimal, evidence-based prompt improvements to prevent agents from attempting GitHub operations outside their scope: PM prompt (4 lines): - Explicit coordination-only boundary - GitHub ops delegation to git-agent - Issue verification requirement Specialist prompts (6 lines each × 13): - Standardized scope boundary section - Do-not checklist (keyword format) - Cross-domain escalation protocol - Pre-work validation checklist Research basis: Perplexity research on prompting best practices found checklist/keyword format achieves 2-3x better compliance than narrative prose with fewer tokens. Files: 14 prompts updated, 97 lines total * feat(#44): expand PM tools and standardize on ripgrep over grep Research-backed expansion of PM tool permissions to enable effective context gathering for informed routing decisions. PM tool additions (read-only): - cat * (all files, not just .project-id) - rg* (ripgrep for fast, gitignore-aware search) - jq* (JSON parsing for GitHub API responses) - wc* (counting and statistics) New instruction file: instructions/tool-preferences.md - System-wide mandate: Use rg over grep - Rationale: faster, respects .gitignore, better defaults - Usage examples and common patterns Research basis: Industry best practice shows orchestrators need read access for informed routing decisions (Perplexity research on multi-agent systems 2024-2025). Files: 2 configs, 1 new instruction, 1 prompt (4 total)
andreipromarketing-dev pushed a commit
to andreipromarketing-dev/opencode
that referenced
this pull request
Apr 7, 2026
Docs: Add hooks.md reference to rules table
xywsxp pushed a commit
to xywsxp/opencode
that referenced
this pull request
Apr 24, 2026
bussard76 pushed a commit
to bussard76/openwork
that referenced
this pull request
May 12, 2026
avion23 pushed a commit
to avion23/opencode
that referenced
this pull request
Jun 10, 2026
6 tasks
zoulukuang added a commit
to zoulukuang/deskfox
that referenced
this pull request
Aug 14, 2026
…] [bug-repro: 点**文件**行后 document.activeElement 又是 body(点**目录**行正常),文件树键盘作用域随之失效 —— 2026-08-13 按 CHECKLIST anomalyco#40 逐项验时实测发现] ## 根因(实测,不是推断) `aaea87be98` 那笔在 handleClick 里加了 `row.focus()`,方向对但**只覆盖了一半**: 点目录行 → 焦点落入文件树 ✅ 点文件行 → 焦点回到 body ❌ 不是「没设焦点」。CDP 实测:点击后该行的 DOM 节点**被销毁重建** (记下点击前的节点引用,点击后 `window.__before !== now` 且 `document.contains(旧节点) === false`)。 触发点是 file-tree.tsx 里这段: inactive={!(local.viewerOpen && local.node.path === local.active)} 文件被打开后该行成为 active 且预览区已开 → 命中 Tooltip 包裹分支 → 行重挂 → 刚设的焦点丢失。 目录行永远不会成为 active,所以不触发 —— 这正是「只修好一半」的由来。 ## 改法 新增 fork-only 纯函数 `file-tree-focus.ts / restoreRowFocus(path)`,在 handleClick 里用 `requestAnimationFrame` 排到重挂之后调用。抽成纯函数是为了能离线回归 —— 这类「重挂丢焦点」靠组件级 e2e 很难稳定复现,而纯函数可以直接构造重挂。 **边界**:只在焦点确实掉回 body 时才补。用户点完文件树又主动点了聊天框,焦点是被有意 转移的,抢回来会比原 bug 更糟 —— 键盘作用域规则是「焦点在哪就归谁」,补焦点不能破坏规则本身。 选择器不用字符串拼接,改遍历比对属性值:文件名含引号会让选择器直接抛 "not a valid selector"(写测试时真撞到),而 `CSS.escape` 并非哪里都有。 ## 验证 - 新增 4 条单测(含 2 条反向:焦点已在别处不许抢、路径不存在不许抛),4/4 通过。 - 真机双向验(重新打包 local 后): 点文件树行 → 焦点停在该行,2.5s 内稳定不变(修复前是 body) 点聊天输入框 → 焦点转移到输入框并保持,**没有被抢回** - app 包单测对照:改动前 870 pass / 12 fail,改动后 874 pass / 12 fail —— 失败项完全相同,系既有基线红,非本笔引入。 - typecheck 通过。 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Wut6R6htBzR8fBN8gbJkWg
zoulukuang added a commit
to zoulukuang/deskfox
that referenced
this pull request
Aug 14, 2026
…diff: 单个新文件,一份不可拆的执行脚本] 新增 `run_group4.py`,覆盖 anomalyco#40~anomalyco#50,全部通过。样本由 `make_fixtures.py` 生成、内容带特征词, 所以「预览对不对」是断言出来的,不是「看着像渲染出来了」。 ## 被实测推翻的「想当然」 - **代码/文本预览渲染在 `<diffs-container>` 的 shadow DOM 里**。 `document.body.innerText` 里根本没有文件内容 —— 于是「打开 plain.txt 后搜不到 TXTMARK」 一度被误读成「代码类文件没渲染」,还去翻了 iframe、canvas、闭合 shadow。 内容一直都在,只是选择器不穿 shadow 边界。已给 uiprobe 加 `deep_find_text` / `selection_text`。 - **对已打开的文件再点一次是「收起预览」**(界面上明写着「点击可收起预览」)。 脚本连开两次同一文件 = 把预览关掉,这是上面那个「搜不到内容」的直接触发条件。 `open_tree_path` 现在会核对预览是否真开着,被收起就再点回来。 - **右键菜单项是普通 `button`,不是 `role=menuitem`**;且**空白处右键给不出菜单** —— 「添加到聊天窗口 / 复制 / 导出为 Word」是针对选区的,必须先拖选。 第一版没选就右键、又按 ARIA role 查,拿到空菜单,差点判成缺陷。 - **`.xlsx` 走 LibreOffice 分页渲染,没有 `<table>`**。第一版按「有没有 table」写, 实际 `tables=0` 却因为文本命中而报了通过 —— 是**断言写松了蒙对**,已改成 断言两个 sheet 的特征词都在(证明整本工作簿都转换了)。 - **`code/` 目录初始就是展开的**,按「先变多再变回」写的断言直接误判(17→12→17); 改成只要求两次点击一开一合、且回到原状,不假设初始态。 ## anomalyco#48 的结论:能力齐全,是条目把入口写死了 右键菜单里**只有「关闭其他标签」**;「关闭标签页」在命令面板/⌘W; 「重新打开已关闭的标签页」命令面板搜不到、`mod+shift+t` 也不响应。 去基准版 `e77443750e`(合上游前)对照:**源码里根本没有 `reopenClosed`** —— 它是上游本次**新增**的命令,注册在 `titlebar.tsx` 的 tab 条上,而 DeskFox 的标签条 不走那条渲染路径。故属「上游新功能未接入」,**不是同步弄丢了功能**,按事实记录不报 FAIL。 CHECKLIST 同步更正 anomalyco#44 / anomalyco#47 / anomalyco#48 三条的预期与工具列。 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Wut6R6htBzR8fBN8gbJkWg
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 freeto join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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.
Description
I was missing clearer instructions on how to get up and running with Bedrock and took the time to write them down in the readme.
I haven't used it but I think we should be able to make this more generic or add information on how to do the same with the supported Azure and GCP models.