Ctrl+P did nothing when the scope held exactly one model and the active
model was a different one. It showed "Only one model in scope" and kept
the current model, because `_cycleScopedModel` returned early for any
list with one entry. `_cycleAvailableModel` had the same early return,
so it also failed when the active model was no longer available and one
other model was.
A second bug had the same cause. When the active model was not in the
list, the code treated it as if it were at index 0, so cycling forward
skipped the first scoped model. With `--models A,B` and another model
active, Ctrl+P went to B.
This change adds a small `getNextCycleIndex` helper that both functions
now use:
- When the active model is not in the list, cycling forward selects the
first entry and cycling backward selects the last.
- It returns `undefined` (the "Only one model…" message) only when the
list is empty, or when its single entry is already the active model.
**Testing**
- Added a regression test to
`test/suite/agent-session-model-extension.test.ts`. It covers a
one-model scope, a repeated cycle once that model is active, and forward
and backward cycling into a two-model scope from outside it. The test
fails without the change ("expected undefined to be 'faux-2'") and
passes with it.
- `npm run check` passes.
- Manual check: `pi --model anthropic/claude-sonnet-4-5 --models
anthropic/claude-opus-4-5`, then Ctrl+P.
- Before: "Only one model in scope".
- After: "Switched to Claude Opus 4.5".
- With `--models anthropic/claude-opus-4-5,anthropic/claude-haiku-4-5`,
Ctrl+P now selects opus instead of haiku.
Fixes#11806
Requested by @davidbrai
Co-authored-by: triagent <triagent@localhost>
Co-authored-by: David Brailovsky <davidbrai@gmail.com>
Fixes#11692.
## Problem
In RPC mode the startup catalog refresh runs in the background and also
refreshes an expired OAuth credential. If pi exits after the token
request went out but before the new token is written, the provider has
already spent the old single-use refresh token and auth.json keeps only
that one. The next request fails with `invalid_grant`.
Trace: `pi --mode rpc` starts, `get_state` is answered, catalog refresh
sends the token request, the client sends SIGTERM (or closes stdin),
`shutdown()` runs `runtimeHost.dispose()`, which does not know about the
refresh, and the process exits. Same for print and interactive mode, and
for a prompt-time refresh cut off by an exit.
## Fix
- `RuntimeCredentials.modify()` tracks every pending modification (same
idiom as the MCP `track()` / `McpAuthProvider.settled()`).
- `RuntimeCredentials.settled()` /
`ModelRuntime.settleCredentialWrites()` wait for them.
- `AgentSessionRuntime.dispose()` waits for them. All modes exit through
`dispose()`. No extra timeout: the OAuth refresh (15 s) and the
auth.json lock wait (30 s) are already bounded, and a second signal
still exits at once.
Not changed: the catalog refresh still refreshes expired OAuth
credentials, also for providers whose catalog does not use them (the
pi.dev catalog for `openai`, `anthropic`, ...). Skipping that would need
a per-provider "catalog needs auth" signal in pi-ai; it is optional now
that exits wait.
`docs/rpc.md` now says that pi waits for pending work before exiting.
## Testing
- `test/runtime-credentials.test.ts`: `settled()` waits until a pending
modification is written, and resolves after a failed one.
-
`test/suite/regressions/11692-dispose-waits-for-oauth-refresh.test.ts`:
`dispose()` waits until the token from a background catalog refresh is
saved. Fails without the fix.
- Real process: `pi --mode rpc` with an extension OAuth provider
(expired credential, 400 ms refresh), SIGTERM about 100 ms after the
token request started. Without the wait: 3 of 3 runs kept the spent
refresh token. With the fix: 3 of 3 saved the new one (exit 143).
Closing stdin instead of SIGTERM: new token saved, exit 0.
- `npm run check`; `test/rpc*`, `test/print-mode`,
`test/model-runtime*`, runtime and regression suites pass.
Fixes#11613.
## Problem
The bash tool waits for stdout/stderr to close, not only for the shell
to exit (#5303). A background process that inherited those pipes and
keeps writing keeps the tool waiting after the shell exited. When the
timeout then hits, the tool reported only `Command timed out after N
seconds`.
Example, timeout 3s:
```bash
bash -c 'while true; do echo tick; sleep 0.05; done' & sleep 1; echo boom >&2; exit 1
```
The shell exits with code 1 after ~1s. The tool still times out at the
deadline, because the loop is still writing to its output. Both facts
are true: the command failed with code 1, and something it started was
still running at the timeout. The old message dropped the first, so the
model read a failed command as a slow one and retried with a longer
timeout, which never helps here.
## Fix
The timeout stays a timeout, since the tool did hit its deadline while
background processes were still writing. When the shell had already
exited before the deadline, the message now also says so:
```
Command timed out after 3 seconds. The shell exited with code 1 after 1.1 seconds, but background processes it started kept writing output until the timeout.
```
A shell still running at the deadline keeps the plain `Command timed out
after N seconds`.
The local backend throws a module-private `ShellTimeoutError` with the
message `timeout:<seconds>` as before, so custom `BashOperations` that
throw `timeout:N` are unaffected.
## Behavior per platform
- Linux and macOS: the shell runs in its own process group (`detached:
true`). At the deadline `kill(-pgid)` still reaches the background
processes after the shell exited, the pipes close, and the tool returns
at the deadline. Tested on Linux; macOS uses the same mechanism and
matches the macOS repros in the issue.
- Windows: still not fully solved. At the deadline `taskkill /T /PID
<shell>` fails with `process not found`, because the shell pid is gone.
Walking parent pids does not help: Git's `bin\bash.exe` is a launcher
for `usr\bin\bash.exe`, and the orphaned loop's parent is that
already-exited inner bash, not the pid pi holds. The tool keeps reading
until the output pauses for 100ms. Measured with Git Bash: the message
is now correct, but the tool returned after 14s instead of 3s; with a
loop that never pauses it never returns. This PR only fixes the
reporting.
## Same pattern elsewhere
- `packages/durable/src/env/node.ts` (`NodeExecutionEnv.exec`): same
code, same false `timeout:N` after the shell exited. Same Windows
limitation.
- `packages/coding-agent/src/core/exec.ts` (`pi.exec` for extensions):
the timeout sends SIGTERM/SIGKILL to the direct child only, no process
group. A descendant that keeps writing keeps the call waiting past the
timeout on every platform, and the result has `killed: true` even when
the child had already exited with its own code.
- `packages/coding-agent/src/core/package-manager.ts` (capture commands
for git/npm): waits for `close` and kills only the direct child on
timeout, so a descendant holding the pipes makes the timeout ineffective
on every platform, even when the descendant writes nothing.
- `packages/mcp/src/transports/stdio.ts`: on Windows, closing a server
that already exited skips taskkill, so children holding its pipes
survive. The code documents this as a known limitation.
Possible follow-up for Windows and the cases above: once the deadline
passed, stop waiting for the pipes (destroy our ends). On Windows this
also ends most writers, which die from a broken pipe (tested with Git
Bash).
## Tests
- `test/suite/regressions/11613-bash-timeout-after-shell-exit.test.ts`
(Unix only; it would hang on Windows for the reason above): background
writer after the shell exits with code 3, and a shell still running at
the deadline. The first case fails without the fix.
- Checked manually on Windows 11 with Git Bash through the built bundle:
both messages are as above.
- `npm run check`, `test/tools.test.ts`, the #5303 and #5208
regressions, `bash-close-hang-windows`, `powershell-tool`,
`shell-kill-process-tree`.
This PR description is AI-generated.
Servers with auth.provider, such as Radius, were reported as needing sign-in because the shell command sent no token. pi mcp now reads and refreshes the provider's /login token, and its hints and login/logout point to /login <provider>.
- type-check subagent-extension.test.ts against the extension view like the other subagent test
- update the test to current ToolRenderContext and ExtensionToolContext types
- use a decodable PNG in the image-only steering test; the fake data was omitted by image processing
## Summary
- remove delivered queue entries even when the user message has no text
- keep image-only steering and follow-up pending counts in sync with the
agent queue
- cover image-only steering delivery with a regression test
## Testing
- `npm run check`
- `./test.sh`
Fixes#8581
Fixes#8409
## Summary
- Pass the request abort signal through lazy stream setup wrappers.
- Report setup failures as aborted when the request signal is already
aborted.
- Added a regression test for aborting during a tool execution before
next auth setup.
Fixes#9124.
Disposing a runtime during tool execution disconnects persistence before
the interrupted turn settles, leaving a stored assistant tool call
without its result. Await `session.abort()` before shutdown and disposal
so the interrupted tool result and final assistant message are
persisted.
This matches `teardownCurrent()`. `AgentSession.dispose()` remains
synchronous. Like the existing replacement path, the wait has no
timeout; handling tools that ignore cancellation is outside this change.
Related: #9986 concerns results for tool calls skipped during a
mid-session abort. This fix preserves the result of an already-running
tool during runtime shutdown.
Validation on main `f1b2e77f5`:
- The regression fails without the fix and passes with it, checking the
durable message order and `Operation aborted` result immediately after
`dispose()` returns, before awaiting the prompt.
- `npm run check`, offline build, and `./test.sh` pass.
Fixes#9221.
An extension tool can lose its context when an RPC command or SDK caller
reloads the session during tool execution: the tool awaits work, reload
invalidates its runner, then `ctx.getSystemPrompt()` throws a
stale-context error. The failed tool result is stored in session
history.
Reject reload when `isStreaming` or `isCompacting` is already set,
before shutdown or runner invalidation. This gives RPC and direct
callers the same entry checks as the TUI. The production change remains
seven lines.
#9548 removed the wrapper access behind the original reproduction. The
regression exercises the tool's own context access after an await,
through both an RPC-style extension command and direct
`AgentSession.reload()`. Rejection preserves the successful result in
session history and the next provider request. Compaction, tree
navigation, and idle reload are also covered.
Scope: this guards operations already marked running or compacting. It
does not serialize reload with prompt preflight or prevent a new prompt
from starting during an ongoing reload.
Related: #5735 proposes a broader deferred-reload API. This patch only
adds the existing TUI entry checks to the core reload method.
Validation on main `f1b2e77f5`:
- Without the guard, four regression cases fail and idle reload passes.
With it, all five pass.
- `npm run check`, offline build, and `./test.sh` pass.
createMcpToolName, mcpNamespace, normalizeMcpInputSchema,
limitMcpContent (with a caller-provided saver) and validation of the
standard mcpServers shape move from coding-agent into pi-mcp, so other
applications name and limit MCP tools exactly like pi. Applications
validate their own fields on the tagged result; coding-agent keeps
exposure, auth.provider and its Client ID Metadata Document rule.
pi-schema: opt-in coercion, compile(schema, { coerce: true }) with coerce() and
parse(), the superset of TypeBox's Value.Convert rules and pi-ai's old coercion;
unresolvedRefs option; url and json-pointer-uri-fragment formats.
pi-ai: Tool takes any JSON Schema; validation through pi-schema; Type, Static
and TSchema are no longer exported (breaking). pi-agent-core and pi-durable:
tools on pi-schema, no TypeBox dependency. Coding agent: extensions keep Type,
Static, TSchema and StringEnum through an extension view of pi-ai; built-in
tools stay on TypeBox.
In sessions created by `createAgentSession`, compaction, branch
summaries and bug-report summaries resolved the stored OAuth credential
in `AgentSession._getSummarizationRequestAuth`, then passed the access
token on as `options.apiKey`. The SDK's stream function calls
`modelRuntime.streamSimple`, which resolves auth again. Because it was
given an explicit `apiKey`, `resolveProviderAuth` took the API-key
override branch instead of using stored OAuth. Built-in providers accept
the token either way. A provider whose API-key path differs from its
OAuth path, such as one that refuses explicit keys, failed every summary
while normal prompts worked.
`createAgentSession` now marks its stream function with
`markRuntimeStreamFunction`. This records it in a module-level `WeakMap`
keyed by the function, mapped to its `ModelRuntime`.
`_getSummarizationRequestAuth` skips pre-resolution when the agent's
stream function is marked for the session's own runtime, so that runtime
resolves auth once, as it does for normal prompts. The check runs after
virtual-model routing. Bare `streamSimple` and custom `streamFn`s still
receive resolved credentials, so the fixes for #6324 and #7579 are
unchanged. If a caller replaces `agent.streamFunction`, the old behavior
applies.
**Testing**
- Added `test/suite/regressions/11520-sdk-summary-oauth-auth.test.ts`.
It builds an SDK session on a provider that has stored OAuth and also an
API-key method, runs a branch summary, and asserts two things: no
`getAuth` call carries an `apiKey` override, and the provider receives
the OAuth access token. It fails without the change and passes with it.
- `npm run check` passes and `./test.sh` passes. I also ran the related
tests: 6324 branch-summary auth, compaction model overrides,
auto-compaction cancellation, SDK stream options, virtual models, and
the compaction and branch-summary tests.
- Manual reproduction: an SDK script with an `openai` OAuth credential
that logs `ModelRuntime.getAuth` during `navigateTree(..., { summarize:
true })`. Before the change there were two calls, the second with
`apiKey = <oauth token>`. After the change there is one call, with no
override.
Fixes#11520
Requested by @badlogic
Co-authored-by: triagent <triagent@localhost>
With `codemode.mode: "on"`, each declared tool gets a description suffix
that says what a script call resolves to. For an object output schema,
`describeOutput` shortens this to field names. Any other schema fell
back to the whole rendered TypeScript type, collapsed onto one line. A
tool with a union output schema therefore put its full output type,
nested fields and comments included, into every request. The reporter's
`process` tool, a union of 7 variants, added about 8,000 characters.
Collapsing the type onto one line also left `//` comments in the middle
of the text.
This change:
- **Unions:** for `anyOf`/`oneOf` schemas where every variant is an
object, the suffix lists each variant's field names, joined by ` | `.
- **`const` fields:** `describeFields` now shows the value of `const`
fields, so a discriminator like `action` stays visible. For example: ``
`{ action: "start", process } | { action: "list", processes, truncated?
}` ``.
- **Other multi-line types**, such as an array of objects: instead of
the inlined type, the suffix says `` a value; `await
describeTool("name")` shows its type ``. `describeTool()` already
returns the full declaration.
- **Single-line types** such as `number` are unchanged.
Testing:
- I added a test in `test/suite/agent-session-codemode.test.ts` that
checks the suffix for three tools: one with a union output, one with an
array-of-objects output, and one with a `number` output. It fails
without the change and passes with it.
- `node ../../node_modules/vitest/dist/cli.js --run
test/suite/agent-session-codemode.test.ts
test/suite/agent-session-mcp.test.ts` passes, including the existing MCP
description assertions.
- `npm run check` is clean.
Fixes#11501
Requested by @badlogic
Co-authored-by: triagent <triagent@localhost>
/model and /scoped-models listed only `id [provider]`, so renamed
overrides from models.json were invisible in the list. Rows now show
the name in parentheses when it has words the id lacks, e.g.
`minimax/minimax-m3 (MiniMax M3 [parasail]) [openrouter]`. Names that
only restate the id with different casing, separators, or a dropped
namespace/version suffix stay hidden.
closes#5916
Pre-prompt recovery of a recoverable length stop omitted the assistant
message but not the tool results that answered its calls, because only
the end-of-run caller passed them. The orphaned results were persisted
and providers rejected every later request.
_omitRecoveryAttempt now finds the matching tool results in the
projected context itself, so every caller omits the pair together.
closes#11227
Extensions importing the public /models subpath (e.g. clampThinkingLevel)
failed with "Cannot find module" because the loader aliased/virtualized
only the pi-ai root, /compat, /oauth, and /providers/all entrypoints.
Add /models to the jiti aliases and the embedded virtual module map,
plus a regression test importing it without a local pi-ai install.
Closes#11156
Co-authored-by: Dante-dan <dantedan@users.noreply.github.com>
Copilot rejects request bodies over its byte-size limit with HTTP 413 and a generic
'failed to parse request' body, so the existing token-count heuristics never matched.
Treat any HTTP 413 as request-size overflow regardless of provider or body text, so
the agent compacts and retries instead of surfacing a raw API error.
Validated against the live Copilot API (40MB image payload): main classifies the 413
as a plain error, this branch detects it as overflow.
Compaction summaries were limited to 0.8 * compaction.reserveTokens (13107 by default), turn-prefix summaries to 0.5 * reserveTokens, and branch summaries to a hardcoded 4096. On providers where reasoning shares the output limit (OpenAI Responses, adaptive Anthropic), high thinking levels used most of that limit and compaction failed with "generation hit the token cap", even though the model allows far more output.
Tying the limit to reserveTokens assumed the summary request is as large as the full context, so its output had to fit in the reserved space. It is usually much smaller: recent messages kept by keepRecentTokens are not sent, a split turn goes to a separate request, tool results are truncated, and tool definitions and the normal system prompt are not included. Manual compaction can also run long before the context is full. The provider layer already sizes output from the actual request, using the model's output limit reduced to fit the remaining context, as for normal turns.
Summary requests now set no maxTokens and rely on that. compaction.reserveTokens only controls when compaction starts.
generateSummary() and generateSummaryWithUsage() now ignore the reserveTokens parameter.
closes#9512, closes#9075, closes#10908
The SDK always passed initial tools, so the constructor never restored the
loadout recorded in the transcript, and resumed sessions dropped tools activated
during the session. Sessions that use the default tools now restore the recorded
loadout, with --tools +name/-name entries applied to it.
fixes#10785
A misplaced options line was ignored, so its timeout_ms never applied and
the script ran without a deadline. parseCodemodeSource now throws
CodemodeSourceError before the script runs.
fixes#10631
joinAdjacentText checked endsWith on the growing joined string, which
makes V8 flatten the concatenation on every item. Collect each text run
in an array, check the separator against the last non-empty part, and
join once. Same fix in the durable code mode copy.
closes#10678
Runs started by pi.sendMessage(..., { triggerTurn: true }) skipped before_agent_start. After their first tool call, the next-turn refresh fell back to the base prompt options and patched away the sections handlers had added, changing the system prompt mid-run and breaking the provider prompt cache.
These runs now go through the same setup as prompt(): compaction check, before_agent_start, tool loadout, nextTurn messages, handler messages, and the system patch. BeforeAgentStartEvent is a union on messageType ("user" or "custom"); custom events carry the starting message.
The session counts as busy from the start of setup, so messages sent while handlers run go into the starting run instead of starting a second one, and an abort requested meanwhile aborts the run once it starts.
closes#10267, closes#5581, closes#10639, closes#10289
partly fixes#10606
Tools that reach other tools declare it with reachesTools(input); MCP uses it to wait for the servers a call needs and to decide which codemode or tool_search tool to activate, instead of parsing codemode scripts and comparing parameter schemas. Namespaces get aliases, replacing codemode's MCP-specific __ suffix rule.
pi-codemode no longer knows MCP: MCP_TYPESCRIPT_PREAMBLE and mcpStructuredContentSchema() are removed, and the MCP result schema describes itself, with the server's local $refs relocated so they resolve as before.
Report idle, working, blocked, done, and error to terminals that support
the Program Status Protocol. ProcessTerminal sends the support query with
the keyboard query and reports only after the terminal answers before DA1;
PI_PROGRAM_STATUS=1|0 overrides detection. Interactive mode reports runs,
compaction, extension dialogs, and logins. agent_settled now carries
aborted so cancelled runs report idle.
closes#10607
The OAuth flow in pi-mcp takes a signal for every request, and an aborted
refresh no longer falls back to a new authorization. The streamable HTTP
transport closes its session with the last request's token instead of
asking the auth provider, which could refresh first.
The MCP extension owns one signal per session: it aborts sign-ins on Esc
and on session shutdown, replaces the generation counter, and shutdown
waits for aborted sign-ins to clean up. Requests to the authorization
server time out after 15s, and pi mcp login --timeout limits the whole
sign-in.
closes#10565
Adds the openai-decisions classifier API with gpt-6-luna on the openai provider. bool questions map to predicates with the criteria in the instructions, refusals fail the result, and gateway timeouts (504) are not retried. The model needs an API key and is hidden from availability while openai uses Sign in with ChatGPT.
Adds optional images to ClassifierContext. Models without image input and APIs that cannot send images return an error result. Codemode's models.classify() accepts images.
Moves the shared classifier HTTP code out of system-one-shared.ts into classifier-shared.ts.
Text items get ==> text N/M <== headers when there are several, and
console output is grouped into one <console_output> block. Console text
items are marked with console: true.
Rules and the tool list follow the declared tools (active minus
hiddenDeclarations). The skills hint names no tool when the reader is
hidden. Codemode shows each tool's prompt guidelines with its
declaration, via the new ToolLoadout.getPromptGuidelines().
closes#10343
--tools no longer removes MCP tools unless an entry starts with mcp__, so
pi --tools codemode can reach MCP servers. Unnamed MCP tools are never
declared directly; only tool_search can load them. --no-tools still
removes everything.
--tools and --exclude-tools accept * patterns, for example
--tools read,codemode,'mcp__radius__*'. --no-mcp disables the built-in
MCP support for one run.
OpenID Connect authorization servers treat a registration without application_type as a web client and reject http loopback redirect URIs. Derive it from redirect_uris when unset (MCP SEP-837): native for loopback hosts and custom schemes, otherwise web.
closes#10493
When the terminal goes away, stdin reads and setRawMode fail with EIO (pi left in an orphaned background process group) or ENOTTY (revoked tty on macOS). These were uncaught, recorded in crashes.json, and prompted users to run /bug. Handle them on stdin like stdout/stderr and skip crash recording for dead terminal errors.
image() output is also written to a temp file and the result names the
path before each image, so later turns can reference generated images.
Output files are now created through one helper (still tmpdir()) and
are readable only by the user.
closes#10310
Add pi.registerToolRenderer((toolName, next) => renderers), which chooses the
renderers for calls to any tool, including tools that are not registered.
Resolvers run in extension load order; next() returns what the remaining
resolvers, then the registered tool, would use. Interactive mode and HTML
export both resolve through it.
The MCP extension uses it to render every mcp__<server>__<tool> call with the
MCP renderers, so calls in a resumed session no longer render fully expanded
before their server connects, or for good if it never does.
closes#10285
Adds oauth.clientRegistration: "cimd", which identifies pi with its Client ID Metadata Document on pi.dev instead of dynamic client registration, matching Codex: a shared document when the authorization server sends iss, otherwise a document and callback path specific to the MCP server. Authorization responses are only accepted on the exact redirect URI. Dynamic client registration stays the default.
closes#10302
Move the codemode script reference, including the models API, into
docs/codemode.md and keep only one line per global in the tool
description. Declared tools get a one-line note on how scripts call them,
and the codemode system prompt guidance and MCP server section are shorter.
With the default tools a GPT-5.6 request drops from about 5,300 to 3,300
tokens.
Errors now point the way back: unknown tools and models members name close
matches, models.classify() and models.generateImages() validate their
arguments, unknown or mistyped models point to getAvailableOfType(),
store() overflow explains the store, and unshown generated images get a note.
Scripts run image models with the session's credentials, like
models.classify(). Results are base64 image blocks that image() attaches
to the codemode result; usage counts toward the session cost.
The session restored its tool loadout before MCP servers reconnected, so deferred tools that tool_search had loaded were dropped. Restored tools that are not registered yet now stay pending and are activated when they register, until a setActiveTools() call deactivates a tool or the next prompt starts.
The configured metadata document replaces discovery and is trusted as
configured. Authorization codes are only exchanged when the response's
iss parameter names the flow's authorization server.
closes#10172
MCP tool and namespace names now replace - with _ like Codex, so the
pi tool name is also the codemode identifier. Tools of a server whose
names collide all get a hash suffix, and server names that differ only
in - and _ are rejected.
closes#10239
In codemode.mode "only", tools whose declarations codemode hides were still
listed in the <tools> section. Drop their snippets so the list matches the
declared tools.
closes#10192
The first prompt waits only for servers with direct tools. Other servers
connect in the background; codemode scripts, tool_search, and the
resource tools wait for the servers they need. codemode and tool_search
are activated from the config.
Tool declarations no longer depend on connected servers: the codemode
description leaves out deferred tools, and the tool_search description
no longer lists sources. Servers are listed in an mcp_servers system
prompt section instead, set in before_agent_start from the config and
connected state, so changes are appended to the conversation.
describeNamespace() and searchTools() accept mcp__dev-radius,
mcp__dev_radius, dev-radius, and dev_radius.
closes#10212
