izackp/takopi
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

docs: specification rewrite

Browse Source
This commit is contained in:
banteg
2026-01-01 21:42:14 +04:00
parent 23f2002836
commit 035441c889
1 changed files with 221 additions and 413 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/specification.md
+221 -413
View File
@@ -1,279 +1,190 @@
# Takopi Specification v0.2.0 [2025-12-31] # Takopi Specification v0.2.0 (minimal) [2025-12-31]
This document specifies Takopi v0.2.0 behavior and architecture in a way that is testable, evolvable, and explicitly aligned with the goals: This document is **normative**. The words **MUST**, **SHOULD**, and **MAY** express requirements.
- **Better testability** ## 1. Scope
- **Runner abstraction** to support future runners (e.g., Claude Code)
- **Telegram remains the only bot client** (adding another is unlikely)
- **Parallel runs are allowed across different threads**, but runs for the **same thread must be serialized** to avoid corrupting history
This is a normative spec using **MUST / SHOULD / MAY** language. Sections labeled **Decision** capture choices that should remain stable unless intentionally changed. Takopi v0.2.0 specifies:
------ - A **Telegram** bot bridge that runs an agent **Runner** and posts:
- a throttled, edited **progress message**
- a **final message** with the final answer and a resume line
- **Thread continuation** via a **resume command** embedded in chat messages
- **Parallel runs across different threads**
- **Serialization within a thread** (no concurrent runs on the same thread)
- A Takopi-owned **normalized event model** produced by runners and consumed by renderers/bridge
## 1. Scope and goals Out of scope for v0.2.0:
### 1.1 Goals (v0.2.0) - Non-Telegram clients (Slack/Discord/etc.)
- Auto-selecting among multiple runners
1. Provide a Telegram bot that runs an “exec agent” (runner) and streams progress updates with periodic edits. - Token-by-token streaming of the assistant’s final answer
2. Support “thread continuation” via a **resume command** embedded in chat messages. - Engines/runners that cannot provide **stable action IDs** within a run
3. Support **parallel execution across different threads** (different resume tokens).
4. Enforce **serialization per thread** (same resume token) to avoid concurrent mutation of the same engine conversation/history.
5. Establish a stable, Takopi-owned **normalized event model** that runners produce and renderers consume.
6. Keep architecture modular enough to add another runner in a future version with minimal changes.
### 1.2 Non-goals (v0.2.0)
- Adding additional bot clients besides Telegram (Discord/Slack/etc.) is out of scope.
- Implementing auto-selection of multiple runners is not required (but should be prepared for).
- Streaming partial assistant answers token-by-token is not required (progress UI is event-driven; final answer is delivered at completion).
- Supporting engines that cannot provide stable action IDs is out of scope (see §5.4).
------
## 2. Terminology ## 2. Terminology
- **Runner / Engine**: Implementation that executes an agent process (Codex today; Claude Code later) and produces Takopi events. - **EngineId**: string identifier of an engine (e.g., `"codex"`).
- **Thread**: The engine-side conversation identifier. In Takopi this is represented as a **ResumeToken**. - **Runner**: Takopi adapter that executes an engine process and yields **Takopi events**.
- **ResumeToken**: A Takopi-owned structured identifier: `{ engine: EngineId, value: str }`. - **Thread**: a single engine-side conversation, identified in Takopi by a **ResumeToken**.
- **ResumeLine**: A runner-owned string representation embedded in chat; **canonical** representation is the engine CLI command (Decision §4.1). - **ResumeToken**: Takopi-owned thread identifier `{ engine: EngineId, value: str }`.
- **Takopi Event**: A normalized event dict emitted by a runner and consumed by renderers/bridge. - **ResumeLine**: a runner-owned string embedded in chat that represents a ResumeToken.
- **Progress Message**: Telegram message that is edited periodically to show live status. - **Run**: a single invocation of `Runner.run(prompt, resume)`.
- **Final Message**: Telegram message containing final answer + resume line + status. - **TakopiEvent**: a normalized event emitted by a runner and consumed by renderers/bridge.
- **Progress message**: a Telegram message that is periodically edited during a run.
- **Final message**: a Telegram message that includes run status, final answer, and resume line.
------ ## 3. Resume tokens and resume lines
## 3. Architecture overview ### 3.1 Decision: canonical resume line is the engine CLI resume command
### 3.1 Layers and responsibilities (strict boundaries) The canonical ResumeLine embedded in chat MUST be the engine’s CLI resume command, e.g.:
**Domain Model (Takopi-owned)** - `codex resume <id>`
- `claude --resume <id>`
- Defines: `ResumeToken`, `TakopiEvent`, `Action` (including the terminal `completed` event). Takopi MUST treat the runner as authoritative for:
- No Telegram, no subprocess, no engine JSON.
**Runner Interface (Takopi-owned)** - formatting a ResumeToken into a ResumeLine
- extracting a ResumeToken from message text
- Defines `Runner` protocol: `run()`, `extract_resume()`, `format_resume()`, etc. ### 3.2 ResumeToken schema (Takopi-owned)
- Runners are trusted producers of Takopi events (Decision §5.2).
**Runner Implementations (engine-owned logic)**
- Codex runner translates engine-specific stream into Takopi events.
- Each runner enforces per-thread serialization (MUST, §6.2).
**Renderers (Takopi-owned)**
- Pure functions/state machines that consume Takopi events and produce markdown strings.
- No engine-specific parsing.
- No Telegram API calls.
**Bridge (Telegram orchestration)**
- Receives Telegram updates and turns them into runner invocations.
- Maintains throttled progress editing.
- Handles cancellation `/cancel`.
- Owns Telegram markdown constraints (limits, entity formatting).
### 3.2 Module naming and one-word modules (v0.2.0 refactor target)
Recommended module layout (single-word filenames, clean layering):
- `takopi/model.py`
Domain types: events, actions, resume token.
- `takopi/runner.py`
Runner protocol.
- `takopi/runners/codex.py`
Codex runner implementation.
- `takopi/runners/mock.py`
Script/mock runner for tests.
- `takopi/render.py`
Progress renderer and event-to-text formatting.
- `takopi/bridge.py`
Telegram orchestration; main loop and message handler.
- `takopi/cli.py`
Typer/CLI entrypoints, config loading, engine selection.
- `takopi/markdown.py`
Markdown sanitization + Telegram entity prep.
**Rationale:**
The normalized event model MUST NOT live under `runners/` because it is core domain state shared by bridge and renderer.
------
## 4. Resume tokens and resume lines
### 4.1 Decision: canonical resume representation is engine CLI command
The canonical representation of “resume” embedded in chat is the runner’s **engine CLI resume command**, e.g.:
- Codex: ``codex resume <uuid>``
- Claude Code: ``claude --resume <uuid>``
Takopi MUST treat the runner as the authority for:
- formatting a `ResumeToken` into a `ResumeLine`
- extracting a `ResumeToken` from message text
Takopi MAY introduce additional Takopi-owned metadata lines in the future (e.g., `resume: codex:<uuid>`), but **v0.2.0 canonical remains the CLI command**.
### 4.2 ResumeToken structure (Takopi-owned)
```python ```python
@dataclass(frozen=True, slots=True) @dataclass(frozen=True, slots=True)
class ResumeToken: class ResumeToken:
engine: str # EngineId (string) engine: str # EngineId
value: str value: str
``` ```
### 4.3 Runner resume codec interface (MUST) ### 3.3 Runner resume codec (MUST)
Each runner MUST implement: Each runner MUST implement:
- `format_resume(token: ResumeToken) -> str` * `format_resume(token: ResumeToken) -> str`
Returns a ResumeLine suitable for embedding in Telegram markdown (usually inside backticks). * `extract_resume(text: str) -> ResumeToken | None`
- `extract_resume(text: str) -> ResumeToken | None` * `is_resume_line(line: str) -> bool`
Extracts a ResumeToken from arbitrary message text.
- `is_resume_line(line: str) -> bool`
Fast check used for truncation safety (to preserve the resume line during trimming).
**Constraints:** Constraints:
- `format_resume()` MUST raise or otherwise fail if `token.engine != runner.engine`. * `format_resume()` MUST fail if `token.engine != runner.engine`.
- `extract_resume()` MUST return `None` if it cannot confidently parse a resume command for its engine. * `extract_resume()` MUST return `None` if it cannot **confidently** parse a resume line for its engine.
### 4.4 Resume extraction behavior in the bridge (v0.2.0) ### 3.4 Bridge resume resolution (MUST)
Given a user message `text` and optional reply-to message `reply_text`: Given `text` (user message) and optional `reply_text` (the message being replied to):
1. The bridge MUST attempt `runner.extract_resume(text)`. 1. The bridge MUST attempt `runner.extract_resume(text)`.
2. If not found, the bridge MUST attempt `runner.extract_resume(reply_text)` if present. 2. If not found, it MUST attempt `runner.extract_resume(reply_text)` if present.
3. If still not found, run starts as a **new thread** (`resume=None`). 3. If still not found, the run MUST start with `resume=None` (new thread).
**Future note (non-normative):** ## 4. Normalized event model
For multi-runner auto-selection, the bridge MAY attempt extraction across all registered runners. This is not required for v0.2.0.
------ ### 4.1 Decision: events are trusted after normalization
## 5. Normalized event model (Takopi-owned) Runners are responsible for emitting well-formed Takopi events. Consumers (renderer/bridge) SHOULD assume validity and MAY fail fast on invariant violations.
### 5.1 Decision: events are trusted after normalization ### 4.2 Supported event types (minimum set)
Runners are responsible for producing well-formed Takopi events. Downstream consumers (render/bridge) SHOULD assume validity and may fail fast if invariants are violated (Decision §5.2). Takopi MUST support:
### 5.2 Event types (minimum set) * `started`
* `action`
* `completed`
Takopi MUST support the following event types: Minimal runner mode is supported:
1. `started` * A runner MAY emit only `started` and `completed`.
2. `action` * If `action` events are emitted, `phase="completed"` alone is valid (no requirement to emit `started`/`updated` phases).
3. `completed`
**Minimal runner mode (supported):** ### 4.3 Event schemas
Runners MAY emit only: All events MUST include `engine: EngineId` and `type`.
- exactly one `started` #### 4.3.1 `started`
- exactly one `completed`
`action` events are optional. If emitted, a runner MAY emit only
`phase="completed"` action events (no requirement to emit `started` / `updated`
phases or track pending action state).
### 5.3 Required fields by event type
#### 5.3.1 `started`
Required: Required:
- `type: "started"` * `type: "started"`
- `engine: EngineId` * `engine: EngineId`
- `resume: ResumeToken` * `resume: ResumeToken`
Optional: Optional:
- `title: str` (human-readable session/agent label) * `title: str`
- `meta: dict` (debug/diagnostic payloads) * `meta: dict`
#### 5.3.2 `action` #### 4.3.2 `action`
Required: Required:
- `type: "action"` * `type: "action"`
- `engine: EngineId` * `engine: EngineId`
- `action: Action` * `action: Action`
- `phase: "started" | "updated" | "completed"` * `phase: "started" | "updated" | "completed"`
Optional: Optional:
- `ok: bool` (typically present when `phase="completed"`) * `ok: bool` (typically on `phase="completed"`)
- `message: str` (freeform status/warning text) * `message: str`
- `level: "debug" | "info" | "warning" | "error"` * `level: "debug" | "info" | "warning" | "error"`
Notes: Notes:
- `phase="completed"` alone is valid; `started` / `updated` are optional. * `phase="completed"` alone is valid.
#### 5.3.3 `completed` #### 4.3.3 `completed`
Required: Required:
- `type: "completed"` * `type: "completed"`
- `engine: EngineId` * `engine: EngineId`
- `ok: bool` (success/failure of the run) * `ok: bool` (overall run success/failure)
- `answer: str` (final assistant response text; may be empty) * `answer: str` (final assistant answer; MAY be empty)
Optional: Optional:
- `resume: ResumeToken` (final resume token for the run; new or existing, if known) * `resume: ResumeToken` (final token; new or existing, if known)
- `error: str | None` (fatal error message, if any) * `error: str | None` (fatal error message, if any)
- `usage: dict` (engine usage/telemetry, if provided) * `usage: dict` (telemetry/usage if available)
### 5.4 Action schema (MUST, per your Decision #4) ### 4.4 Action schema (MUST; stable IDs)
Actions MUST have stable IDs. Actions MUST have stable IDs within a run:
```python ```python
@dataclass(frozen=True, slots=True) @dataclass(frozen=True, slots=True)
class Action: class Action:
id: str # required id: str
kind: str # required, stable taxonomy kind: str
title: str # required, short label title: str
detail: dict[str, Any] # required, structured details detail: dict[str, Any]
``` ```
**Definition (v0.2.0):** Stability requirements:
“Stable” means **stable within a single run**: the same underlying action MUST keep the same `Action.id` across all events in that run, and `Action.id` values MUST be unique within the run. Takopi does not require action IDs to remain stable across different runs/resumes.
Action kinds SHOULD be from a stable set (extensible): * Within a single run, the same underlying action MUST keep the same `Action.id` across events.
* `Action.id` values MUST be unique within a run.
* IDs do **not** need to be stable across different runs/resumes.
- `command` Action kinds SHOULD come from an extensible stable set, e.g.:
- `tool`
- `file_change`
- `web_search`
- `note`
- `turn`
- `warning`
- `telemetry`
- `note`
Runners MAY include additional kinds, but renderers MAY treat unknown kinds as `note`. * `command`, `tool`, `file_change`, `web_search`, `turn`, `warning`, `telemetry`, `note`
The `detail` dict is **freeform per runner**; no per-kind schema is enforced. Renderers SHOULD handle missing or unexpected fields gracefully. Unknown kinds MAY be rendered as `note`.
The `ok` field semantics are **runner-defined**. For example, a runner MAY treat `grep` exit code 1 (no match) as `ok=True` if contextually appropriate. `detail` is freeform; no per-kind schema is required.
**User-visible warnings and errors:** runners SHOULD surface these as `action` events with `phase="completed"` (typically `kind="warning"` or `kind="note"`) and `ok=False`, rather than introducing additional event types. `ok` semantics are runner-defined.
------ User-visible warnings/errors SHOULD be surfaced as `action` events (typically `kind="warning"` or `kind="note"`, `phase="completed"`, `ok=False`) rather than introducing new event types.
## 6. Runner interface and concurrency semantics ## 5. Runner protocol and concurrency
### 6.1 Runner protocol (MUST) ### 5.1 Runner protocol (MUST)
```python ```python
class Runner(Protocol): class Runner(Protocol):
engine: str engine: str # EngineId
def run( def run(
self, self,
@@ -282,288 +193,185 @@ class Runner(Protocol):
) -> AsyncIterator[TakopiEvent]: ... ) -> AsyncIterator[TakopiEvent]: ...
``` ```
### 6.2 Per-thread serialization (MUST; core invariant) ### 5.2 Per-thread serialization (MUST; core invariant)
**Invariant:** At most one active run may operate on the same thread (same `ResumeToken`) at a time. Define:
- Parallel runs are allowed only if they target **different** threads. * `ThreadKey(resume) := f"{resume.engine}:{resume.value}"`
- Runs targeting the same thread MUST be queued and executed sequentially.
- This invariant MUST be enforced by the runner implementation (even if used outside the bridge).
**Critical requirement for new sessions:** Invariant:
If `resume is None`, the runner MUST acquire the per-thread lock **as soon as the new thread's ResumeToken becomes known**, and MUST do so **before emitting `started`** to downstream consumers.
This prevents: * At most **one** active run may operate on the same `ThreadKey` at a time.
- a second run resuming the thread while the original "new session" run is still active Rules:
- history corruption due to concurrent engine operations
**Bridge note (non-normative):** * Runs for different ThreadKeys MAY run in parallel.
The bridge may enforce FIFO scheduling per thread to avoid emitting multiple progress messages for the same thread while a run is already in-flight. * Runs for the same ThreadKey MUST be queued and executed sequentially.
* This invariant MUST be enforced by the runner implementation even if used outside the Telegram bridge.
**Codex note (non-normative):** New thread rule (`resume is None`):
Codex emits `thread.started` (with `thread_id`) before any `turn.*` / `item.*` events for both new and resumed runs. Codex MAY emit top-level warning `error` lines (e.g., config warnings) before `thread.started`; the Codex runner translates these warnings into `action` events with `phase="completed"` and yields them in the same order as received (so `started` is not guaranteed to be the first yielded event). If the subprocess exits before `thread.started` is observed, no `started` can be emitted and the bridge reports an error without a resume line.
Codex also emits exactly one `agent_message`/`assistant_message` per turn; the runner uses that message text as `completed.answer`. * When the runner learns the new thread’s ResumeToken, it MUST:
**Claude Code note (non-normative):** * acquire the per-thread lock for that token
Claude Code emits `system.init` (with `session_id`) before any `assistant`/`user` message objects; the runner should emit `started` on `system.init`. Claude’s final `result` message carries the session id and final answer (`result.result`), which the runner uses as `completed.answer`. * do so **before emitting** `started(resume=token)`
### 6.3 Run completion event (MUST) ### 5.3 `started` emission and ordering
```python * If the runner obtains a ResumeToken for the run, it MUST emit exactly one `started` event containing that token.
@dataclass(frozen=True, slots=True) * The runner MAY emit `action` events before `started` (e.g., pre-init warnings). Consumers MUST NOT assume `started` is the first event.
class CompletedEvent:
type: Literal["completed"]
engine: EngineId
ok: bool # success/failure of the run
resume: ResumeToken | None = None # final resume token for the run (new or existing, if known)
answer: str # final assistant response text (may be empty)
```
`completed` MUST be the final event of a successful run. ### 5.4 Completion
### 6.4 Event delivery semantics (MUST) * If the run reaches `started`, and then terminates under the runner’s control (success or detected failure), the runner MUST emit exactly one `completed` event and it MUST be the last event.
* If the runner never obtains a ResumeToken (e.g., fatal failure before session init), it MAY emit no `started` and no `completed`.
Event ordering is significant. The system MUST ensure: ### 5.5 Event delivery semantics (MUST)
- Events are yielded to the consumer in the same order they are produced by the runner. * Events MUST be yielded in the order produced by the runner.
- Event delivery MUST NOT spawn unbounded background tasks per event. * The runner MUST NOT spawn unbounded background tasks per event.
- If the consumer stops iteration early (break/cancel/exception), the runner MUST abort the run (best-effort) and release any held resources. * If the consumer stops iterating early (cancel/break/exception), the runner MUST abort the run best-effort and release any held locks/resources.
### 6.5 Crash and error handling ## 6. Bridge (Telegram orchestration)
If the runner subprocess crashes or exits uncleanly: ### 6.1 Responsibilities (MUST)
- The bridge MUST publish an error status message.
- If `started` was received, the bridge MUST include the resume line in the error message.
------
## 7. Bridge (Telegram orchestration)
### 7.1 Responsibilities
The bridge MUST: The bridge MUST:
- Poll Telegram updates. * Receive Telegram updates
- Resolve resume token (from message text or reply target). * Resolve resume token (per §3.4)
- Start runner execution with appropriate cancellation support. * Schedule runs per thread (per §6.2)
- Maintain progress rendering and Telegram edits (rate-limited). * Start runner execution with cancellation support
- Publish final answer and include resume line. * Maintain a progress message with rate-limited edits
- Support `/cancel` to cancel the run associated with an in-flight progress message. * Publish a final message containing status, answer, and resume line (when known)
* Support `/cancel` for in-flight runs
**Queuing behavior:**
- Multiple prompts to the same thread are queued and executed sequentially.
- There is no queue depth limit; all prompts are accepted.
### 7.1.1 Scheduling algorithm (MUST)
The bridge MUST implement per-thread FIFO scheduling in a way that does not require spawning one task per queued job.
**Definitions:**
- `ThreadKey := f"{resume.engine}:{resume.value}"`
- `Job := (chat_id, user_msg_id, text, resume: ResumeToken | None)`
**Required behavior:**
- For `resume != None`, the bridge MUST enqueue the job into `pending_by_thread[ThreadKey]` and ensure exactly one worker drains that queue sequentially.
- If a run starts with `resume == None` but later emits `started(resume=token)`, the bridge MUST treat that run as the in-flight job for `ThreadKey(token)` for scheduling purposes until it completes.
- A thread worker MUST exit when its queue is empty; the bridge SHOULD avoid retaining per-thread state for inactive threads.
The bridge MUST NOT: The bridge MUST NOT:
- parse engine-native events * parse engine-native streams/events
- encode engine-specific rules beyond resume extraction via runner * embed engine-specific rules beyond calling runner resume extraction/formatting
### 7.2 Progress behavior Queue depth:
- The bridge SHOULD send an initial progress message quickly (“running…”). * There is no queue depth limit; all prompts are accepted.
- The bridge SHOULD edit the progress message no more frequently than every 2 seconds.
- The bridge SHOULD avoid edits if rendered content has not changed.
### 7.3 Resume line inclusion ### 6.2 Scheduling (MUST)
The progress renderer and/or final message MUST include the canonical resume line once known: Definitions:
- If `started` has been received, the progress view SHOULD include the resume line. * `Job := (chat_id, user_msg_id, text, resume: ResumeToken | None)`
- The final message MUST include the resume line.
**Important:** because the resume line may appear during progress updates, the bridge MUST treat `started` as the point at which the thread key becomes known for scheduling and cancellation routing. Required behavior:
### 7.4 Cancellation `/cancel` * For `resume != None`, the bridge MUST enqueue jobs into `pending_by_thread[ThreadKey(resume)]`.
* For each ThreadKey, exactly one worker (or equivalent mechanism) MUST drain the queue sequentially.
* A worker MUST exit when its queue is empty; the bridge SHOULD avoid retaining state for inactive threads.
* The implementation MUST avoid spawning one long-lived task per queued job (bounded concurrency).
- The bridge MUST allow the user to cancel a run in progress by sending `/cancel` in reply to the progress message (or by other defined mapping). Runs that start as new threads:
- Cancel MUST terminate the runner process via **SIGTERM** and stop further progress edits.
- After cancellation, the bridge MUST publish a "cancelled" status message and SHOULD include the resume line if known.
- If `/cancel` is sent with additional text, the additional text is ignored; only cancellation occurs.
### 7.5 Telegram markdown constraints * If a job starts with `resume=None` and later yields `started(resume=token)`, the bridge MUST treat that run as the in-flight job for `ThreadKey(token)` until it completes (for scheduling and cancellation routing).
### 6.3 Progress message behavior
* The bridge SHOULD send an initial progress message quickly (e.g., “Running…”).
* The bridge SHOULD edit the progress message no more frequently than every **2 seconds**.
* The bridge SHOULD skip edits when rendered content is unchanged.
* Once `started` is observed, the progress view SHOULD include the canonical ResumeLine.
### 6.4 Final message requirements (MUST)
The final output MUST include:
* a status line (`done` / `error` / `cancelled`)
* the final `answer` (if any)
* the ResumeLine if known (and MUST include it if `started` was received)
### 6.5 Cancellation `/cancel` (MUST)
* The bridge MUST allow users to cancel a run in progress by sending `/cancel` in reply to the progress message (or by an equivalent mapping defined by the bridge).
* Cancellation MUST terminate the runner process via **SIGTERM**.
* After cancellation, the bridge MUST stop further progress edits and publish a “cancelled” status message.
* The bridge SHOULD include the ResumeLine if known.
* Any additional text after `/cancel` is ignored.
### 6.6 Telegram markdown + truncation (MUST)
The bridge MUST: The bridge MUST:
- escape/prepare markdown per Telegram rules * escape/prepare Telegram markdown correctly
- enforce Telegram message length limits (including after escaping) * enforce Telegram message length limits (including after escaping)
- avoid truncating away the resume line (use runner `is_resume_line()`) * avoid truncating away the ResumeLine (using `runner.is_resume_line()`)
If truncation is required: If truncation is required:
- the bridge MUST keep the resume line intact * the bridge MUST keep the ResumeLine intact
- the bridge SHOULD preserve the **head** (beginning) of content and add an ellipsis marker before truncation point * the bridge SHOULD preserve the beginning of the content and insert an ellipsis at the truncation point
------ ### 6.7 Crash/error handling (MUST)
## 8. Renderer (progress and final formatting) If the runner crashes or exits uncleanly:
### 8.1 Renderer responsibilities * the bridge MUST publish an error status message
* if `started` was received, the bridge MUST include the ResumeLine in that error message
## 7. Renderer
Renderers MUST: Renderers MUST:
- be deterministic functions of Takopi events and internal state * be deterministic functions/state machines over Takopi events + internal renderer state
- produce markdown text and (optionally) entity annotations * produce Telegram-ready markdown (or markdown + entities)
* tolerate `action` events that are “completed-only” (no prior `started`/`updated`)
Renderers MUST NOT: Renderers MUST NOT:
- depend on engine-native events * depend on engine-native event formats
- call Telegram APIs * call Telegram APIs
- perform blocking operations * perform blocking I/O
### 8.2 Progress renderer state Action update collapsing:
The progress renderer SHOULD maintain: * If multiple `action` events share the same `Action.id`, renderers SHOULD treat later `started`/`updated` events as updates (replace the prior running line rather than appending).
- session title ## 8. Configuration and engine selection
- current running actions and their latest summaries
- completed actions and status
- resume token if known
The progress renderer MUST tolerate “completed-only” actions (no prior Decision (v0.2.0):
`started` / `updated`) and treat them as standalone steps.
If the runner emits multiple `action` events for the same `Action.id` while it is still running (e.g., repeated `phase="started"` or `phase="updated"`), the progress renderer SHOULD treat these as updates and collapse them into a single line (replacing the prior running line rather than appending a new one). * Exactly one runner is selected at startup via a CLI subcommand (no default).
* If no engine subcommand is provided, Takopi prints an engine chooser panel and exits.
* Resume extraction uses only the selected runner.
* If a user provides a resume line for a different engine, extraction fails and the bridge treats the message as a new thread (`resume=None`).
### 8.3 Final rendering ## 9. Testing requirements (MUST)
Final output MUST include: Tests MUST cover:
- status line (`done` / `error` / `cancelled`) 1. **Runner contract**
- final `answer`
- resume line
------ * If a token is obtained: exactly one `started`
* Action schema validity (required fields; stable unique IDs within run)
* Event ordering preserved
* `completed` emitted and last for controlled termination after `started`
2. **Runner serialization**
## 9. Configuration and engine selection * Concurrent runs for the same ResumeToken serialize
* `resume=None` runs acquire the per-thread lock once token is known and before emitting `started`
3. **Bridge per-thread scheduling**
### 9.1 v0.2.0 behavior (Decision #5) * FIFO per ThreadKey
* second job for same thread does not start until first completes
4. **Progress throttling**
- A single runner/engine is selected at startup via CLI subcommand (no default). * edits not more frequent than configured interval
- If no engine subcommand is provided, Takopi prints the engine chooser panel and exits. * no edit when content unchanged
- Resume extraction uses only the selected runner’s parser. * truncation preserves ResumeLine
- If the user attempts to resume a thread created by a different engine, resume extraction will fail and the bot treats it as a new thread. 5. **Cancellation**
### 9.2 Future behavior (non-normative) * `/cancel` terminates run and produces “cancelled”
* ResumeLine included if known
6. **Renderer formatting**
Takopi MAY support: * completed-only actions render correctly
* repeated events for same Action.id collapse as intended
- trying all registered runners’ `extract_resume` to auto-select a runner for resumes Test tooling SHOULD include event factories, deterministic/fake time, and a script/mock runner.
- selecting a preferred engine from config when no resume is present
The architecture SHOULD keep this future change localized to a `RunnerRegistry` / router.
------
## 10. Testing requirements (v0.2.0)
### 10.1 Test categories (MUST)
1. **Runner contract tests**
- Emits exactly one `started`
- All actions (if any) have required fields and stable IDs
- `completed.resume` matches started token (when present)
- Event ordering is preserved
- `ok` semantics match intended behavior
2. **Runner serialization tests (critical)**
- Serializes concurrent runs for the same `ResumeToken`
- For `resume=None`, acquires per-thread lock once the token is known (before emitting `started`)
3. **Bridge per-thread scheduling tests (critical)**
- Enqueue two prompts for the same `ResumeToken`
- Assert the bridge does not start the second run until the first completes
4. **Bridge progress throttling tests**
- Edits no more frequently than configured interval
- No edits without changes
- Truncation preserves resume line
5. **Cancellation tests**
- `/cancel` terminates run
- “cancelled” status produced
- resume line included if known
6. **Renderer formatting tests**
- Correct rendering of actions
- Stable formatting under event sequences
### 10.2 Test tooling guidelines (SHOULD)
- Provide **event factories** in tests for readability.
- Provide a deterministic fake clock/sleep.
- Use a script/mock runner to simulate event sequences.
------
## 11. Open design notes / evolution hooks
### 11.1 Takopi-owned resume tags (future discussion)
Even though canonical is engine CLI command in v0.2.0, Takopi MAY later add a Takopi-owned unambiguous line such as:
- `resume: codex:<uuid>`
Benefits:
- easier multi-runner routing
- resilience to CLI syntax changes
- simpler truncation and extraction
This is not required for v0.2.0.
### 11.2 EngineId typing
To reduce friction adding new runners, v0.2.0 SHOULD treat engine IDs as strings (or a `NewType(str)`), not a closed Literal union.
------
## 12. Changelog template (for evolving this spec)
- v0.2.0 [2025-12-31]
- Establish Takopi normalized event model and runner protocol
- Canonical resume representation is engine CLI command
- Enforce per-thread serialization including new sessions once token is known
- Telegram-only bridge with progress edits + cancellation
- Recommended module split into one-word modules
- Clarify: `ok` semantics are runner-defined, `detail` is freeform
- Clarify: bridge queues per thread (FIFO)
- Clarify: SIGTERM for cancellation, `/cancel` ignores accompanying text
- Clarify: truncation preserves head + resume line
- Clarify: crash publishes error with resume if known
------
## Appendix A: Example end-to-end flow (informative)
1. User sends: “Refactor this module and run tests.”
2. Bridge resolves resume token:
- none in message, none in reply → `resume=None`
3. Bridge sends a progress message: “Running…”
4. Runner starts and emits:
- `started(engine="codex", resume={engine:"codex", value:"<uuid>"})`
- `action(id="1", kind="command", title="pytest", detail={...}, phase="started")`
- `action(id="1", ok=True, phase="completed", ...)`
- `completed(resume=..., ok=True, answer="...")`
5. Progress renderer now includes resume line:
- ``codex resume <uuid>``
6. User replies to progress message with follow-up prompt.
7. Bridge extracts resume via runner, chooses same thread, and queues it behind the in-flight run if still active.
8. Final message includes:
- “done”
- final answer
- resume line ``codex resume <uuid>``