benchflow 0.7.2.dev1986__tar.gz → 0.7.2.dev1987__tar.gz

{benchflow-0.7.2.dev1986 → benchflow-0.7.2.dev1987}/CHANGELOG.md

@@ -14,6 +14,20 @@
   undetectable repos fall back silently to the path-derived source id.
 
 ### Changed
+- **The contributor prompt now tells the agent to upgrade BenchFlow first.**
+  `CONTRIBUTOR_PROMPT` (kept in sync in `README.md`, `docs/traj-upload.md`,
+  and the `benchflow-traj-upload` skill evals) is a three-line block that
+  says to run `uv tool install --python 3.12 --upgrade --force benchflow`
+  before reading the skill, so agents that skim the skill or hit a stale copy
+  still install the latest CLI. The prompt also names OpenCode and Cursor
+  sessions and the re:Agent hackathon 72-hour window; README/docs render it
+  as a blockquote (soft-wraps on GitHub) behind an explicit "send this to
+  your coding agent" framing, and `bench traj setup` prints the same framing.
+  The skill's Discover step gains best-effort Cursor
+  (`~/.cursor/projects/*/agent-transcripts/`) and OpenCode
+  (`~/.local/share/opencode/opencode.db`, `opencode db path`) locations plus
+  a prefer-recent-matching-sessions note. Follow-up to the version
+  precondition from #1013/#1014.
 - **Trajectory viewer restyled to match www.benchflow.ai.** All three
   `bench eval view` pages (stream-json/JSONL, ACP events, multi-turn trial)
   now share one inline stylesheet with the site's design language: light

{benchflow-0.7.2.dev1986 → benchflow-0.7.2.dev1987}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: benchflow
-Version: 0.7.2.dev1986
+Version: 0.7.2.dev1987
 Summary: Multi-turn agent benchmarking with ACP — run any agent, any model, any provider.
 Project-URL: Homepage, https://github.com/benchflow-ai/benchflow
 Project-URL: Repository, https://github.com/benchflow-ai/benchflow
@@ -89,20 +89,23 @@ BenchFlow is a universal environment framework: it runs AI agents against task e
 
 ## Quick start: 1. Submit a trajectory
 
-Proud of a Claude Code or Codex session? Copy this line into Claude Code,
-Codex, Cursor, or any coding agent. That is the $2,000 eval prize path. No
-BenchFlow account, API key, or Azure login.
+**Don't follow these steps yourself — send them to your coding agent.** Copy
+the block below and paste it as a message to Claude Code, Codex, OpenCode, or
+Cursor. That is the $2,000 eval prize path. No BenchFlow account, API key, or
+Azure login.
 
-```
-Submit my best local Claude Code or Codex session to the BenchFlow eval prize. Read https://raw.githubusercontent.com/benchflow-ai/benchflow/main/.agents/skills/benchflow-traj-upload/SKILL.md and follow it: find a session, open the viewer, upload only after I review it.
-```
+> Submit my relevant local Claude Code, Codex, OpenCode, or Cursor session from the re:Agent e2e agentic science / ai4bio hackathon (last 72 hours, on this laptop) to the BenchFlow eval prize.
+>
+> 1. First make sure the latest benchflow CLI is installed: uv tool install --python 3.12 --upgrade --force benchflow
+>
+> 2. Then read https://raw.githubusercontent.com/benchflow-ai/benchflow/main/.agents/skills/benchflow-traj-upload/SKILL.md and follow it: find a session, open the viewer, and upload only after I (the human) review it.
 
 The agent finds sessions on this machine, opens the viewer, and uploads after
 you like what you see.
 
-Be on the latest BenchFlow first — `uv tool install --python 3.12 --upgrade
---force benchflow`. The `bench traj` commands print a one-line upgrade hint
-when a newer release is available.
+The prompt has the agent install the latest BenchFlow first (`uv tool
+install --python 3.12 --upgrade --force benchflow`). The `bench traj` commands
+also print a one-line upgrade hint when a newer release is available.
 
 Prefer the terminal instead? The guided upload inspects before anything leaves
 your machine — it renders a redacted trajectory report (step counts, masked
@@ -128,7 +131,7 @@ bench traj setup
 ```
 
 `npx skills add` asks which agents to install for. `bench traj setup` copies
-the skill into this project and prints the same paste line. See the
+the skill into this project and prints the same agent prompt. See the
 [upload skill](./.agents/skills/benchflow-traj-upload/SKILL.md).
 
 ## Quick start: 2. Run with a ChatGPT or Claude subscription

{benchflow-0.7.2.dev1986 → benchflow-0.7.2.dev1987}/README.md

@@ -15,20 +15,23 @@ BenchFlow is a universal environment framework: it runs AI agents against task e
 
 ## Quick start: 1. Submit a trajectory
 
-Proud of a Claude Code or Codex session? Copy this line into Claude Code,
-Codex, Cursor, or any coding agent. That is the $2,000 eval prize path. No
-BenchFlow account, API key, or Azure login.
+**Don't follow these steps yourself — send them to your coding agent.** Copy
+the block below and paste it as a message to Claude Code, Codex, OpenCode, or
+Cursor. That is the $2,000 eval prize path. No BenchFlow account, API key, or
+Azure login.
 
-```
-Submit my best local Claude Code or Codex session to the BenchFlow eval prize. Read https://raw.githubusercontent.com/benchflow-ai/benchflow/main/.agents/skills/benchflow-traj-upload/SKILL.md and follow it: find a session, open the viewer, upload only after I review it.
-```
+> Submit my relevant local Claude Code, Codex, OpenCode, or Cursor session from the re:Agent e2e agentic science / ai4bio hackathon (last 72 hours, on this laptop) to the BenchFlow eval prize.
+>
+> 1. First make sure the latest benchflow CLI is installed: uv tool install --python 3.12 --upgrade --force benchflow
+>
+> 2. Then read https://raw.githubusercontent.com/benchflow-ai/benchflow/main/.agents/skills/benchflow-traj-upload/SKILL.md and follow it: find a session, open the viewer, and upload only after I (the human) review it.
 
 The agent finds sessions on this machine, opens the viewer, and uploads after
 you like what you see.
 
-Be on the latest BenchFlow first — `uv tool install --python 3.12 --upgrade
---force benchflow`. The `bench traj` commands print a one-line upgrade hint
-when a newer release is available.
+The prompt has the agent install the latest BenchFlow first (`uv tool
+install --python 3.12 --upgrade --force benchflow`). The `bench traj` commands
+also print a one-line upgrade hint when a newer release is available.
 
 Prefer the terminal instead? The guided upload inspects before anything leaves
 your machine — it renders a redacted trajectory report (step counts, masked
@@ -54,7 +57,7 @@ bench traj setup
 ```
 
 `npx skills add` asks which agents to install for. `bench traj setup` copies
-the skill into this project and prints the same paste line. See the
+the skill into this project and prints the same agent prompt. See the
 [upload skill](./.agents/skills/benchflow-traj-upload/SKILL.md).
 
 ## Quick start: 2. Run with a ChatGPT or Claude subscription

{benchflow-0.7.2.dev1986 → benchflow-0.7.2.dev1987}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "benchflow"
-version = "0.7.2.dev1986"
+version = "0.7.2.dev1987"
 description = "Multi-turn agent benchmarking with ACP — run any agent, any model, any provider."
 readme = "README.md"
 requires-python = ">=3.12"

{benchflow-0.7.2.dev1986 → benchflow-0.7.2.dev1987}/src/benchflow/cli/traj.py

@@ -54,14 +54,26 @@ SKILL_RAW_URL = (
     "/.agents/skills/benchflow-traj-upload/SKILL.md"
 )
 
+UPGRADE_COMMAND = "uv tool install --python 3.12 --upgrade --force benchflow"
+
+# Three unwrapped lines separated by blank lines: each logical unit is one
+# full physical line (no hard wraps mid-sentence), so the URL and the upgrade
+# command stay selectable in agents and terminals.
 CONTRIBUTOR_PROMPT = (
-    "Submit my best local Claude Code or Codex session to the BenchFlow eval "
-    "prize. Read "
-    f"{SKILL_RAW_URL} "
-    "and follow it: find a session, open the viewer, upload only after I review it."
+    "Submit my relevant local Claude Code, Codex, OpenCode, or Cursor session "
+    "from the re:Agent e2e agentic science / ai4bio hackathon (last 72 hours, "
+    "on this laptop) to the BenchFlow eval prize."
+    "\n\n"
+    f"1. First make sure the latest benchflow CLI is installed: {UPGRADE_COMMAND}"
+    "\n\n"
+    f"2. Then read {SKILL_RAW_URL} "
+    "and follow it: find a session, open the viewer, and upload only after "
+    "I (the human) review it."
 )
 
-UPGRADE_COMMAND = "uv tool install --python 3.12 --upgrade --force benchflow"
+CONTRIBUTOR_PROMPT_FRAMING = (
+    "Send this to your coding agent (it's a prompt for the agent, not steps for you):"
+)
 
 
 def _fetch_latest_version() -> str | None:
@@ -166,14 +178,14 @@ def register_traj(app: typer.Typer) -> None:
         ] = False,
         prompt_only: Annotated[
             bool,
-            typer.Option("--prompt", help="Print the copy-paste agent line and exit"),
+            typer.Option("--prompt", help="Print the copy-paste agent prompt and exit"),
         ] = False,
         list_sessions: Annotated[
             bool,
             typer.Option("--list", help="List recent local sessions and exit"),
         ] = False,
     ) -> None:
-        """Install the submit skill, or print the line to paste into an agent."""
+        """Install the submit skill, or print the prompt to send to an agent."""
         _maybe_print_update_hint()
         if prompt_only:
             _print_contributor_prompt()
@@ -194,7 +206,6 @@ def register_traj(app: typer.Typer) -> None:
                 _run_npx_skill_install()
         else:
             _install_project_skill(Path.cwd())
-        console.print("Paste this to your agent:")
         _print_contributor_prompt()
         if interactive and typer.confirm(
             "List recent sessions and open the viewer now?", default=False
@@ -625,7 +636,9 @@ def _print_dry_run(staged: StagedCapture) -> None:
 
 
 def _print_contributor_prompt() -> None:
-    # One physical line so people can copy it. Rich wrapping would break that.
+    # Plain print: Rich wrapping would break the unbroken URL line and make
+    # the block awkward to copy.
+    print(CONTRIBUTOR_PROMPT_FRAMING)
     print(CONTRIBUTOR_PROMPT)
 
 
@@ -671,7 +684,7 @@ def _run_npx_skill_install() -> None:
         check=False,
     )
     if completed.returncode != 0:
-        print_error("npx skills add failed; the copy-paste line below still works.")
+        print_error("npx skills add failed; the copy-paste prompt below still works.")
 
 
 def _print_session_hits() -> None:

{benchflow-0.7.2.dev1986 → benchflow-0.7.2.dev1987}/tests/test_traj_upload_cli.py

@@ -715,17 +715,45 @@ def test_skip_update_check_env_var_never_fetches(
 
 
 def test_setup_prompt_prints_the_copy_paste_line() -> None:
-    """The human path is one line to paste into an agent."""
-    from benchflow.cli.traj import CONTRIBUTOR_PROMPT, SKILL_RAW_URL
+    """The human path is one prompt block to send to an agent.
+
+    Guards the version step and hand-off framing added in PR #1017
+    (follow-up to PRs #1013 and #1014): the prompt tells the agent to
+    upgrade BenchFlow before reading the skill, the upgrade command appears
+    before the skill URL, and the URL sits on one unbroken physical line.
+    README.md renders the prompt as a blockquote (a fenced code block would
+    not soft-wrap on GitHub), so each unwrapped prompt line is asserted to be
+    present in the README instead of the whole string verbatim.
+    """
+    from benchflow.cli.traj import (
+        CONTRIBUTOR_PROMPT,
+        CONTRIBUTOR_PROMPT_FRAMING,
+        SKILL_RAW_URL,
+        UPGRADE_COMMAND,
+    )
 
     result = runner.invoke(app, ["traj", "setup", "--prompt"])
     readme = Path(__file__).resolve().parents[1] / "README.md"
+    readme_text = readme.read_text(encoding="utf-8")
 
     assert result.exit_code == 0, result.output
-    assert CONTRIBUTOR_PROMPT in click.unstyle(result.output)
+    output = click.unstyle(result.output)
+    assert CONTRIBUTOR_PROMPT_FRAMING in output
+    assert CONTRIBUTOR_PROMPT in output
     assert SKILL_RAW_URL in result.output
     assert "bench traj upload" not in result.output
-    assert CONTRIBUTOR_PROMPT in readme.read_text(encoding="utf-8")
+
+    prompt_lines = [line for line in CONTRIBUTOR_PROMPT.splitlines() if line]
+    assert len(prompt_lines) == 3
+    for line in prompt_lines:
+        assert line in readme_text
+    assert "send them to your coding agent" in readme_text
+
+    assert UPGRADE_COMMAND in CONTRIBUTOR_PROMPT
+    assert CONTRIBUTOR_PROMPT.index(UPGRADE_COMMAND) < CONTRIBUTOR_PROMPT.index(
+        SKILL_RAW_URL
+    )
+    assert any(SKILL_RAW_URL in line for line in prompt_lines)
 
 
 def test_setup_yes_installs_the_skill(
@@ -740,7 +768,9 @@ def test_setup_yes_installs_the_skill(
     assert skill.is_file()
     text = skill.read_text(encoding="utf-8")
     assert "open the viewer" in text
-    assert "Paste this to your agent" in click.unstyle(result.output)
+    # PR #1017 reframed the hand-off copy: the block is a prompt for the
+    # agent, not steps for the human.
+    assert "Send this to your coding agent" in click.unstyle(result.output)
 
 
 def test_list_recent_sessions_scans_all_projects_most_recent_first(