fix(tmux-context): contain executor rejection as a warning, correct suppression claim

Review round on #758.

bash.run() only promises to resolve for nonzero exits, timeouts, and aborts,
and bash.resolve() can reject on policy grounds, so either could escape the
serial agent/step listener and abort the model turn — contradicting the
plugin's documented failed-query no-op contract. Contain both and log a
warning instead; the location is optional context.

The Agent Note claimed an unchanged location suppresses the query. It does
not: only the interval floor is checked before the query, while change
suppression compares state the query returned. Corrected in both languages
and re-recorded the i18n pairs.
This commit is contained in:
Turtle
2026-07-29 11:11:17 +08:00
parent f15ff737cd
commit 44a657c149
9 changed files with 73 additions and 17 deletions

View File

@@ -2,5 +2,5 @@
# side as of the last confirmed-consistent state. Both languages carry equal authority;
# after editing either side, bring the other along and re-record with:
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-27-tmux-location-context.md
2026-07-27-tmux-location-context.md: 9436e9bcf764a505a4da3c8f5ef1d6313ba648d3
2026-07-27-tmux-location-context.zh.md: 09cbf056a577918bc8f9682843154b8dc1bda5a0
2026-07-27-tmux-location-context.md: bac5861f7f55c259de04d153115f164d90c415ad
2026-07-27-tmux-location-context.zh.md: 03cd722381c45604f7aae8f3d0a9f9fbb8b12bb5

View File

@@ -42,11 +42,11 @@ The published `./invariant` companion registers no runtime check: a reading is a
## Consequences
An agent booted inside tmux now receives its own session/window/pane location and window layout as durable, source-attributed context, updated per turn when the location changes. Deployments opt in through cordis.yml; the default spine and shipped examples stay silent. Outside a real tmux pane — including a terminal that merely inherited `$TMUX`/`$TMUX_PANE` — or without a `ctx.bash` executor, the plugin is inert with no error, so composing it is safe everywhere. Because the reading is one durable `user/message`, it survives compaction as ordinary history, contributes nothing to system-prompt assembly or request headers, and costs at most one two-line message per changed turn. The pull model adds one `tmux display-message` subprocess (through the sandboxed bash seam) on the first step of each turn that is due; unchanged locations and the optional interval floor suppress both the query and the injection.
An agent booted inside tmux now receives its own session/window/pane location and window layout as durable, source-attributed context, updated per turn when the location changes. Deployments opt in through cordis.yml; the default spine and shipped examples stay silent. Outside a real tmux pane — including a terminal that merely inherited `$TMUX`/`$TMUX_PANE` — or without a `ctx.bash` executor, the plugin is inert with no error, so composing it is safe everywhere. Because the reading is one durable `user/message`, it survives compaction as ordinary history, contributes nothing to system-prompt assembly or request headers, and costs at most one two-line message per changed turn. The pull model adds one `tmux display-message` subprocess (through the sandboxed bash seam) on the first step of each turn that is due. The optional interval floor is checked before the query and so suppresses both; an unchanged location is detected only by comparing the returned state, so it suppresses the injection while still paying for the query.
## Testing
Unit tests pin: first-step injection and source/surface metadata; the `$TMUX_PANE`-keyed command including its `#{pane_tty}`-vs-`ps -o tty=` guard; step-gating; change suppression across turns and re-injection on a moved pane; positive-interval suppression and threshold; every no-op path (no bash, nonzero exit, wrong field count, empty pane id, aborted signal); prepended ordering before ordinary `agent/step` listeners; resilience to a corrupt prior reading (non-text block, single-line text); and config rejection of negative and non-integer intervals. Per-file coverage is 100%.
Unit tests pin: first-step injection and source/surface metadata; the `$TMUX_PANE`-keyed command including its `#{pane_tty}`-vs-`ps -o tty=` guard; step-gating; change suppression across turns and re-injection on a moved pane; positive-interval suppression and threshold; every no-op path (no bash, nonzero exit, wrong field count, empty pane id, aborted signal, and a contained executor rejection from either `resolve()` or `run()` that warns instead of failing the turn); prepended ordering before ordinary `agent/step` listeners; resilience to a corrupt prior reading (non-text block, single-line text); and config rejection of negative and non-integer intervals. Per-file coverage is 100%.
## Alternatives considered

View File

@@ -42,11 +42,11 @@ window active=<0|1>, pane active=<0|1>, layout <window-layout>
## 后果
启动于 tmux 内的 agent 现在会以持久、带来源标记的上下文收到自身的 session/window/pane 位置及 window 布局,并在位置变化时按轮次更新。部署方通过 cordis.yml 选择启用;默认 spine 与随附示例保持沉默。在真实 tmux pane 之外——包括仅继承了 `$TMUX``$TMUX_PANE` 的终端——或没有 `ctx.bash` 执行器时,插件保持惰性且不报错,因此在任何地方组合它都安全。由于读数是一条持久的 `user/message`,它作为普通历史经受压缩,对系统提示装配与请求头毫无贡献,且每个发生变化的轮次至多花费一条两行消息。拉取模型在每个到期轮次的第一个 step 增加一次 `tmux display-message` 子进程(经沙箱化的 bash seam;位置未变化以及可选的间隔下限会同时抑制查询与注入
启动于 tmux 内的 agent 现在会以持久、带来源标记的上下文收到自身的 session/window/pane 位置及 window 布局,并在位置变化时按轮次更新。部署方通过 cordis.yml 选择启用;默认 spine 与随附示例保持沉默。在真实 tmux pane 之外——包括仅继承了 `$TMUX``$TMUX_PANE` 的终端——或没有 `ctx.bash` 执行器时,插件保持惰性且不报错,因此在任何地方组合它都安全。由于读数是一条持久的 `user/message`,它作为普通历史经受压缩,对系统提示装配与请求头毫无贡献,且每个发生变化的轮次至多花费一条两行消息。拉取模型在每个到期轮次的第一个 step 增加一次 `tmux display-message` 子进程(经沙箱化的 bash seam。可选的间隔下限在查询之前检查,因此同时抑制查询与注入;而位置未变化只能通过比较查询返回的状态得知,因此它只抑制注入,查询开销仍会付出
## 测试
单元测试固定了:首个 step 的注入及来源/表层元数据;以 `$TMUX_PANE` 为键的命令(含其 `#{pane_tty}``ps -o tty=` 的比对守卫step 门槛;跨轮次的变化抑制与 pane 移动时的重新注入;正间隔抑制与阈值;每条空操作路径(无 bash、非零退出、字段数不符、pane id 为空、信号已取消);前置排序先于普通 `agent/step` 监听器;对损坏的历史读数(非文本块、单行文本)的容错;以及配置对负值与非整数间隔的拒绝。逐文件覆盖率为 100%。
单元测试固定了:首个 step 的注入及来源/表层元数据;以 `$TMUX_PANE` 为键的命令(含其 `#{pane_tty}``ps -o tty=` 的比对守卫step 门槛;跨轮次的变化抑制与 pane 移动时的重新注入;正间隔抑制与阈值;每条空操作路径(无 bash、非零退出、字段数不符、pane id 为空、信号已取消,以及 `resolve()``run()` 抛出的执行器拒绝被兜住并记录警告而非使该轮失败);前置排序先于普通 `agent/step` 监听器;对损坏的历史读数(非文本块、单行文本)的容错;以及配置对负值与非整数间隔的拒绝。逐文件覆盖率为 100%。
## 考虑过的替代方案

View File

@@ -1499,7 +1499,7 @@ export interface Config {
}
```
Source: [`packages/context/tmux-context/src/index.ts:33`](../packages/context/tmux-context/src/index.ts)
Source: [`packages/context/tmux-context/src/index.ts:34`](../packages/context/tmux-context/src/index.ts)
## `@deepseek-ai/dsh-token-meter`

View File

@@ -2,5 +2,5 @@
# side as of the last confirmed-consistent state. Both languages carry equal authority;
# after editing either side, bring the other along and re-record with:
# pnpm run verify-translation-pairing --write packages/context/tmux-context/README.md
README.md: 5ea36948d6d83135c5aa97650c0d77e942adbbaa
README.zh.md: 914d8d7c99c37de2c64541bcf4968996d819077d
README.md: a166a46d20f472cb5d8f045e2456ce3e6de7a2f2
README.zh.md: 0575d549e352239e7d954870eaf40beea1169cc6

View File

@@ -27,7 +27,7 @@ pane_tty=$(tmux display-message -t "$TMUX_PANE" -p '#{pane_tty}') || exit 1
exec tmux display-message -t "$TMUX_PANE" -p '<format>'
```
`$TMUX_PANE` alone is insufficient: a terminal launched from a tmux shell (a VS Code integrated terminal, a desktop launcher) **inherits** `$TMUX` and `$TMUX_PANE` from that ancestor, so the variables are present even though the process does not live in that pane. The command therefore also compares the pane's `#{pane_tty}` against this process's own controlling terminal (`ps -o tty=` for its pid): a genuine pane owns this process's tty, while an inherited environment names some other pane's tty. Running through `ctx.bash` applies the deployment's sandbox and policy; the plugin owns no subprocess code. When `ctx.bash` is absent, the process is not in a real tmux pane (`$TMUX_PANE` unset, or the tty does not match ⇒ nonzero exit), or the reading is malformed, the attempt is a no-op, never an error.
`$TMUX_PANE` alone is insufficient: a terminal launched from a tmux shell (a VS Code integrated terminal, a desktop launcher) **inherits** `$TMUX` and `$TMUX_PANE` from that ancestor, so the variables are present even though the process does not live in that pane. The command therefore also compares the pane's `#{pane_tty}` against this process's own controlling terminal (`ps -o tty=` for its pid): a genuine pane owns this process's tty, while an inherited environment names some other pane's tty. Running through `ctx.bash` applies the deployment's sandbox and policy; the plugin owns no subprocess code. When `ctx.bash` is absent, the process is not in a real tmux pane (`$TMUX_PANE` unset, or the tty does not match ⇒ nonzero exit), or the reading is malformed, the attempt is a no-op, never an error. The location is optional, so an executor rejection — a policy refusal from `resolve()` or an infrastructure failure from `run()` — is contained and logged as a warning rather than failing the turn.
State is pulled on every eligible turn — a moved, renamed, or re-laid-out pane is picked up without any tmux hook or background process. The plugin re-injects only when the rendered tmux state differs from its last injection, so an unchanged location adds nothing.

View File

@@ -27,7 +27,7 @@ pane_tty=$(tmux display-message -t "$TMUX_PANE" -p '#{pane_tty}') || exit 1
exec tmux display-message -t "$TMUX_PANE" -p '<format>'
```
仅凭 `$TMUX_PANE` 并不足够:从 tmux shell 启动的终端VS Code 集成终端、桌面启动器)会从该祖先进程**继承** `$TMUX``$TMUX_PANE`,因此即使进程并不位于那个 pane 中,这些变量依然存在。为此该命令还会把 pane 的 `#{pane_tty}` 与本进程自己的控制终端(对其 pid 执行 `ps -o tty=`)作比较:真正的 pane 拥有本进程的 tty而继承而来的环境指向的是另一个 pane 的 tty。通过 `ctx.bash` 运行会应用部署方的沙箱与策略;插件不拥有任何子进程代码。当 `ctx.bash` 缺失、进程不在真实的 tmux pane 内(`$TMUX_PANE` 未设置,或 tty 不匹配 ⇒ 非零退出)或读取结果格式非法时,本次尝试为空操作,绝不报错。
仅凭 `$TMUX_PANE` 并不足够:从 tmux shell 启动的终端VS Code 集成终端、桌面启动器)会从该祖先进程**继承** `$TMUX``$TMUX_PANE`,因此即使进程并不位于那个 pane 中,这些变量依然存在。为此该命令还会把 pane 的 `#{pane_tty}` 与本进程自己的控制终端(对其 pid 执行 `ps -o tty=`)作比较:真正的 pane 拥有本进程的 tty而继承而来的环境指向的是另一个 pane 的 tty。通过 `ctx.bash` 运行会应用部署方的沙箱与策略;插件不拥有任何子进程代码。当 `ctx.bash` 缺失、进程不在真实的 tmux pane 内(`$TMUX_PANE` 未设置,或 tty 不匹配 ⇒ 非零退出)或读取结果格式非法时,本次尝试为空操作,绝不报错。由于位置信息是可选的,执行器的拒绝——`resolve()` 的策略拒绝或 `run()` 的基础设施故障——会被兜住并记录为警告,而不会使该轮失败。
状态在每个符合条件的轮次拉取——pane 被移动、改名或重新布局都会被感知,无需任何 tmux hook 或后台进程。插件仅在渲染出的 tmux 状态与上次注入不同时才重新注入,因此位置不变时不会新增任何内容。

View File

@@ -12,15 +12,16 @@
* only when the rendered tmux state changes since the last injection (a moved,
* renamed, or re-laid-out pane), with an optional `refreshIntervalMs` floor
* between injections. Absent tmux environment, an inherited-only environment,
* absent `ctx.bash`, or a failed query is a no-op, never an error.
* absent `ctx.bash`, or a failed query is a no-op, never an error: an executor
* rejection is contained and logged as a warning so the turn continues.
*
* @module @deepseek-ai/dsh-tmux-context
*/
import type { Context } from 'cordis'
import type { Context, LoggerService } from 'cordis'
import z from 'schemastery'
import type { Agent } from '@deepseek-ai/dsh-agent'
import type { BashExecutor } from '@deepseek-ai/dsh-bash'
import type { BashExecutor, BashRunResult } from '@deepseek-ai/dsh-bash'
import { createUserMessage } from '@deepseek-ai/dsh-llm'
/** Cordis plugin name used by loader diagnostics. */
@@ -92,13 +93,20 @@ const FIELD_SEP = '\\t'
* on a match, so an inherited environment reads as "not in tmux" and injects
* nothing.
*
* The location is optional context, so an executor rejection is a failed query,
* not a turn failure: `resolve()` may reject the command on policy grounds and
* `run()` only promises to resolve for nonzero exits, timeouts, and aborts, so
* both are contained and reported as a warning.
*
* @param bash - the executor seam used to run the read-only tmux/ps commands.
* @param logger - receives a warning when the executor rejects the query.
* @param processId - this agent process's pid, whose controlling tty must match the pane.
* @param signal - abort signal forwarded to the executor.
* @returns the parsed location, or `undefined` when not in a real pane or on any failure.
*/
async function queryTmuxLocation(
bash: BashExecutor,
logger: LoggerService,
processId: number,
signal: AbortSignal,
): Promise<TmuxLocation | undefined> {
@@ -111,8 +119,14 @@ async function queryTmuxLocation(
'[ "$pane_tty" = "/dev/$self_tty" ] || exit 1',
`exec tmux display-message -t "$TMUX_PANE" -p '${format}'`,
].join('\n')
const spec = bash.resolve({ command, signal })
const result = await bash.run(spec)
let result: BashRunResult
try {
result = await bash.run(bash.resolve({ command, signal }))
} catch (error: unknown) {
const message = error instanceof Error ? error.message : String(error)
logger.warn(`tmux location query failed: ${message}; injecting no location this turn`)
return undefined
}
if (result.exitCode !== 0) return undefined
const line = result.stdout.text.split('\n', 1)[0] as string
const parts = line.split(FIELD_SEP)
@@ -215,7 +229,7 @@ export function apply(ctx: Context, config: Config): void {
const now = Date.now()
if (now >= previous.time && now - previous.time < refreshIntervalMs) return
}
const location = await queryTmuxLocation(bash, process.pid, signal)
const location = await queryTmuxLocation(bash, ctx.logger, process.pid, signal)
if (location === undefined) return
const state = renderState(location)
if (previous !== undefined && previous.state === state) return

View File

@@ -51,8 +51,10 @@ class FakeBash extends BashExecutor {
commands: string[] = []
result: BashRunResult = runResult(`${tmuxLine()}\n`)
runError?: Error
resolveError?: Error
override resolve(request: BashExecRequest): BashExecSpec {
if (this.resolveError) throw this.resolveError
return {
command: request.command,
workdir: request.workdir ?? '/work',
@@ -325,6 +327,46 @@ describe('tmux-context no-op paths', () => {
expect(contextTexts(session)).toHaveLength(0)
})
it('warns and injects nothing when the executor rejects the run', async () => {
const { ctx, bash } = await mount({}, true)
bash.runError = new Error('bash executor unavailable')
const warn = vi.spyOn(ctx.logger, 'warn')
const session = new Session(SessionId('run-rejected'))
openMessageTurn(session, 1)
await fire(ctx, sessionAgent(session), 1, 1)
expect(contextTexts(session)).toHaveLength(0)
expect(warn).toHaveBeenCalledWith(expect.stringContaining('bash executor unavailable'))
})
it('warns and injects nothing when the executor rejects the command at resolve', async () => {
const { ctx, bash } = await mount({}, true)
bash.resolveError = new Error('command denied by policy')
const warn = vi.spyOn(ctx.logger, 'warn')
const session = new Session(SessionId('resolve-rejected'))
openMessageTurn(session, 1)
await fire(ctx, sessionAgent(session), 1, 1)
expect(contextTexts(session)).toHaveLength(0)
expect(warn).toHaveBeenCalledWith(expect.stringContaining('command denied by policy'))
})
it('reports a non-Error rejection in the warning', async () => {
const { ctx, bash } = await mount({}, true)
// Non-Error throw: the executor seam is typed, but a bad impl can reject with anything.
bash.runError = 'spawn refused' as unknown as Error
const warn = vi.spyOn(ctx.logger, 'warn')
const session = new Session(SessionId('non-error-rejection'))
openMessageTurn(session, 1)
await fire(ctx, sessionAgent(session), 1, 1)
expect(contextTexts(session)).toHaveLength(0)
expect(warn).toHaveBeenCalledWith(expect.stringContaining('spawn refused'))
})
it('skips an already-aborted step and runs before ordinary agent/step listeners', async () => {
const { ctx } = await mount({}, true)
const session = new Session(SessionId('ordering'))