Pine/deepseek-harness
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases 1 Wiki Activity

fix(examples): isolate headless agent example

Browse Source
This commit is contained in:
Tianyi Cui
2026-07-16 16:12:27 +08:00
parent ae73ef774d
commit 8d35092e26
38 changed files with 410 additions and 140 deletions
Download Patch File Download Diff File Expand all files Collapse all files
examples/coding-agent/README.md
+2 -16
View File
@@ -1,6 +1,6 @@
# coding-agent
Coding-agent demo wiring: DeepSeek V4 + the `read`/`write`/`edit` filesystem tools + the bash tool suite + subagent delegation + workflows + `todo_write` + JSONL persistence. `cordis.yml` runs the terminal readline REPL; `cli.cordis.yml` keeps the same coding capabilities behind a headless one-shot CLI.
Coding-agent REPL wiring: DeepSeek V4 + the `read`/`write`/`edit` filesystem tools + the bash tool suite + subagent delegation + workflows + `todo_write` + JSONL persistence.
## Run it
@@ -21,20 +21,6 @@ Type a coding task. The agent works through the `read`/`write`/`edit` filesystem
…
```
### One-shot CLI
Run one task through all model and tool steps, flush its fresh session, print the final result, and exit:
```sh
pnpm run demo:headless -- "fix the failing test in this workspace"
pnpm run demo:headless --output-format json -- "summarize the current implementation"
pnpm run demo:headless --output-format stream-json -- "run the focused tests"
```
The root command supplies `cli.cordis.yml`, which disables HMR and the REPL app and inserts [`@deepseek-ai/dsh-cli-demo`](../../packages/examples/cli-demo). Exactly one quoted positional task is required; there is no `-p` flag. `text` prints the last text-bearing assistant message, `json` prints one DSH-native result record, and `stream-json` emits the parent `main` session's canonical task-turn events before that record. Non-completed turns retain partial output but exit nonzero; argument and boot failures leave stdout empty.
This is non-interactive automation with the same local bash, filesystem, skill, subagent, workflow, and todo capabilities as the REPL. It can mutate the launch workspace and spend provider tokens. No prompt, approval, resume, further turn, or stdin context is available in v1; see the [CLI package contract](../../packages/examples/cli-demo/README.md).
### Resuming a prior session
Each run starts a fresh session by default (its event log lands under `./.sessions/`). To **continue** a previous conversation, set `RESUME_SESSION_ID` to that session's id — the `main` agent then rehydrates the persisted log instead of starting fresh, so the model sees the earlier turns as history:
@@ -83,4 +69,4 @@ This example is a thin leaf `cordis.yml`: it picks the swappable backends, loads
- `tests/compaction.e2e.ts` — the compaction smoke: a real multi-step bash task runs with a deliberately tiny context window so the auto-compaction listener fires MID-SESSION. Verifies the WORLD — a `compact/start…end` pair landed in the real log, the surface shrank (a replace node shadowed older nodes), and the agent still produced a correct final answer after compaction.
- `tests/todo-write.e2e.ts` — a real model drives the real `todo_write` tool and the test verifies the resulting `todo/write` session event.
These self-skip without `DEEPSEEK_API_KEY`. `tests/code-mode.e2e.ts` is the with-key Code Mode proof — a real model, a two-tool task, asserting the wire tool list was exactly `[run_code]`, the `tool/code-dispatch` events landed under the parent call, and the curated answer came back. `tests/cli.e2e.ts` runs the one-shot bin with a real model and verifies its temporary file externally. The keyless Loader smokes run in the default e2e gate: `tests/keyless-smoke.e2e.ts`, `tests/code-mode-keyless-smoke.e2e.ts`, and `tests/cli-keyless-smoke.e2e.ts`; the CLI smoke mocks only the LLM boundary and asserts a real bash round trip plus persisted stream output.
These self-skip without `DEEPSEEK_API_KEY`. `tests/code-mode.e2e.ts` is the with-key Code Mode proof — a real model, a two-tool task, asserting the wire tool list was exactly `[run_code]`, the `tool/code-dispatch` events landed under the parent call, and the curated answer came back. The keyless Loader smokes run in the default e2e gate: `tests/keyless-smoke.e2e.ts` and `tests/code-mode-keyless-smoke.e2e.ts`.