flyloft 0.1.0__py3-none-any.whl

flyloft/_version.py

@@ -0,0 +1,24 @@
+# file generated by vcs-versioning
+# don't change, don't track in version control
+from __future__ import annotations
+
+__all__ = [
+    "__version__",
+    "__version_tuple__",
+    "version",
+    "version_tuple",
+    "__commit_id__",
+    "commit_id",
+]
+
+version: str
+__version__: str
+__version_tuple__: tuple[int | str, ...]
+version_tuple: tuple[int | str, ...]
+commit_id: str | None
+__commit_id__: str | None
+
+__version__ = version = '0.1.0'
+__version_tuple__ = version_tuple = (0, 1, 0)
+
+__commit_id__ = commit_id = None

flyloft/cli/main.py

@@ -0,0 +1,196 @@
+import getpass
+import json
+import os
+from pathlib import Path
+from uuid import UUID, uuid4
+
+import typer
+
+from flyloft.config.load import Config, load_campaign_file
+from flyloft.record import actions, store
+from flyloft.record.db import connect
+from flyloft.record.migrate import apply_pending
+from flyloft.record.models import SteeringBody
+from flyloft.record.views import campaign_views
+from flyloft.turns.roles import ROLES
+from flyloft.workspace import git as ws
+
+app = typer.Typer(no_args_is_help=True, help="flyloft: agent swarms for empirical research on Slurm clusters.")
+campaign_app = typer.Typer(no_args_is_help=True, help="Start, inspect, and steer campaigns.")
+run_app = typer.Typer(no_args_is_help=True, help="Act on runs.")
+db_app = typer.Typer(no_args_is_help=True, help="Database maintenance.")
+app.add_typer(campaign_app, name="campaign")
+app.add_typer(run_app, name="run")
+app.add_typer(db_app, name="db")
+
+
+def _user() -> str:
+    return os.environ.get("USER") or getpass.getuser()
+
+
+def _open():
+    cfg = Config.load()
+    return cfg, connect(cfg.root.database_url)
+
+
+@db_app.command("migrate")
+def db_migrate():
+    """Apply pending schema migrations."""
+    _, conn = _open()
+    applied = apply_pending(conn)
+    typer.echo(f"applied: {applied or 'nothing'}")
+
+
+@app.command()
+def serve():
+    """Run the service loop in the foreground."""
+    from flyloft.service.factory import build_service
+    from flyloft.service.tick import run_forever
+
+    cfg = Config.load()
+    service = build_service(cfg)
+    service.startup()
+    run_forever(service, cfg.root.tick_interval)
+
+
+@campaign_app.command("start")
+def campaign_start(file: Path):
+    """Create a campaign from a campaign file, one agent per role, and post its start entry."""
+    cfg, conn = _open()
+    cf = load_campaign_file(file)
+    try:
+        cfg.validate_campaign(cf)
+    except ValueError as e:
+        typer.echo(str(e), err=True)
+        raise typer.Exit(code=2)
+    project = cfg.projects[cf.project]
+    campaign_id = uuid4()
+    branch = f"flyloft/{campaign_id.hex[:8]}"
+    repo = ws.ensure_clone(cfg.root.data_dir, project)
+    ws.create_branch(repo, project.base_branch, branch)
+    agents = []
+    for name, role in ROLES.items():
+        agent_id = uuid4()
+        worktree = cfg.root.data_dir / "campaigns" / str(campaign_id) / "worktrees" / str(agent_id)
+        ws.add_worktree(repo, branch, worktree, detach=not role.edits_code)
+        agents.append((agent_id, name, cf.roles[name], worktree))
+    with conn.transaction():
+        store.create_campaign(
+            conn,
+            id=campaign_id,
+            project=cf.project,
+            cluster=cf.cluster or project.cluster,
+            question=cf.question,
+            criterion=cf.criterion,
+            baseline=cf.baseline,
+            constraints=cf.constraints,
+            submission_cap=cf.submission_cap,
+            branch=branch,
+        )
+        for agent_id, name, selection, worktree in agents:
+            store.create_agent(
+                conn,
+                id=agent_id,
+                campaign_id=campaign_id,
+                role=name,
+                substrate=selection.harness,
+                model=selection.model or cfg.harnesses[selection.harness].model,
+                workspace_path=str(worktree),
+            )
+        actions.steer(
+            conn,
+            campaign_id=campaign_id,
+            author=_user(),
+            summary="Campaign started",
+            body=SteeringBody(type="start", text=f"Campaign started by {_user()} from {file}"),
+        )
+    typer.echo(str(campaign_id))
+
+
+@campaign_app.command("list")
+def campaign_list():
+    """List campaigns with their state and their agents' states."""
+    _, conn = _open()
+    for c in store.list_campaigns(conn):
+        agents = ", ".join(f"{a.role}:{a.state}" for a in store.agents_for_campaign(conn, c.id))
+        typer.echo(f"{c.id}  {c.project:<12} {c.state:<8} {agents}  {c.question[:50]}")
+
+
+@campaign_app.command("show")
+def campaign_show(campaign_id: UUID):
+    """Hypotheses, runs, findings, pull request, and submissions for one campaign."""
+    _, conn = _open()
+    c = store.get_campaign(conn, campaign_id)
+    v = campaign_views(conn, c.id)
+    typer.echo(f"Campaign {c.id} [{c.state}]" + (f"  blocked: {c.block_reason}" if c.block_reason else ""))
+    typer.echo(f"Question:  {c.question}\nCriterion: {c.criterion}\nBaseline:  {c.baseline}")
+    typer.echo(f"Branch: {c.branch}  PR: {c.pr_url or '(none)'}")
+    typer.echo(f"Submissions: {len(v.runs)} of {c.submission_cap}")
+    typer.echo("\nHypotheses:")
+    for h in v.hypotheses:
+        typer.echo(f"  {h.id} [{h.state}] {h.title}: {h.change}")
+    typer.echo("\nRuns:")
+    for r in v.runs:
+        seen = r.last_observed_at.strftime("%H:%M:%S") if r.last_observed_at else "never"
+        typer.echo(f"  {r.id} {r.purpose:<8} {r.commit_sha[:10]} [{r.state}] seen {seen} {r.metrics or ''}")
+    typer.echo("\nFindings:")
+    for f in v.findings:
+        typer.echo(f"  {f.id} [{f.state}] {f.claim}  (hypothesis {f.hypothesis_id})")
+
+
+@campaign_app.command("log")
+def campaign_log(
+    campaign_id: UUID, full: bool = typer.Option(False, "--full", help="Print bodies and assignment paths.")
+):
+    """Entries and turns interleaved by time."""
+    _, conn = _open()
+    items = []
+    for e in store.entries_for_campaign(conn, campaign_id):
+        line = f"{e.created_at:%Y-%m-%d %H:%M:%S} [{e.kind}] {e.author_kind}:{e.author[:8]} {e.summary}"
+        if full:
+            line += "\n    " + json.dumps(e.body)
+        items.append((e.created_at, line))
+    for t in store.turns_for_campaign(conn, campaign_id):
+        line = f"{t.started_at:%Y-%m-%d %H:%M:%S} [turn] {t.role} via {t.substrate} exit={t.exit_status}"
+        if full:
+            line += f"\n    assignment: {t.assignment_path}"
+        items.append((t.started_at, line))
+    for _, line in sorted(items, key=lambda x: x[0]):
+        typer.echo(line)
+
+
+@campaign_app.command("steer")
+def campaign_steer(campaign_id: UUID, text: str):
+    """Post an instruction. A blocked campaign becomes active."""
+    _, conn = _open()
+    actions.steer(
+        conn,
+        campaign_id=campaign_id,
+        author=_user(),
+        summary=text[:80],
+        body=SteeringBody(type="instruction", text=text),
+    )
+    typer.echo("posted")
+
+
+@campaign_app.command("stop")
+def campaign_stop(campaign_id: UUID):
+    """Stop new work. Outstanding runs finish; open hypotheses are abandoned."""
+    _, conn = _open()
+    actions.steer(conn, campaign_id=campaign_id, author=_user(), summary="Stop", body=SteeringBody(type="stop"))
+    typer.echo("stopped")
+
+
+@run_app.command("cancel")
+def run_cancel(run_id: UUID):
+    """Cancel a run. The service issues scancel on its next tick."""
+    _, conn = _open()
+    run = store.get_run(conn, run_id)
+    actions.steer(
+        conn,
+        campaign_id=run.campaign_id,
+        author=_user(),
+        summary=f"Cancel run {run_id}",
+        body=SteeringBody(type="cancel_run", run_id=run_id),
+    )
+    typer.echo("cancel requested")

flyloft/config/load.py

@@ -0,0 +1,61 @@
+import os
+import tomllib
+from dataclasses import dataclass
+from pathlib import Path
+
+from flyloft.turns.roles import ROLES
+
+from .models import CampaignFile, ClusterConfig, FlyloftConfig, HarnessConfig, ProjectConfig
+
+
+def _read(path: Path) -> dict:
+    with path.open("rb") as f:
+        return tomllib.load(f)
+
+
+def _named(dir: Path, model):
+    out = {}
+    if dir.is_dir():
+        for p in sorted(dir.glob("*.toml")):
+            out[p.stem] = model(name=p.stem, **_read(p))
+    return out
+
+
+@dataclass
+class Config:
+    config_dir: Path
+    root: FlyloftConfig
+    projects: dict[str, ProjectConfig]
+    clusters: dict[str, ClusterConfig]
+    harnesses: dict[str, HarnessConfig]
+
+    @classmethod
+    def load(cls, config_dir: Path | None = None) -> "Config":
+        config_dir = Path(config_dir or os.environ.get("FLYLOFT_CONFIG_DIR", Path.home() / ".config" / "flyloft"))
+        root_data = _read(config_dir / "flyloft.toml")
+        return cls(
+            config_dir=config_dir,
+            root=FlyloftConfig(**root_data),
+            projects=_named(config_dir / "projects", ProjectConfig),
+            clusters=_named(config_dir / "clusters", ClusterConfig),
+            harnesses=_named(config_dir / "harnesses", HarnessConfig),
+        )
+
+    def validate_campaign(self, cf: CampaignFile) -> None:
+        if cf.project not in self.projects:
+            raise ValueError(f"unknown project '{cf.project}'")
+        for role in cf.roles:
+            if role not in ROLES:
+                raise ValueError(f"unknown role '{role}'")
+        for role in ROLES:
+            if role not in cf.roles:
+                raise ValueError(f"campaign file needs a [roles.{role}] table")
+            if cf.roles[role].harness not in self.harnesses:
+                raise ValueError(f"unknown harness '{cf.roles[role].harness}' for role '{role}'")
+        cluster = cf.cluster or self.projects[cf.project].cluster
+        if cluster not in self.clusters:
+            raise ValueError(f"unknown cluster '{cluster}'")
+
+
+def load_campaign_file(path: Path) -> CampaignFile:
+    return CampaignFile(**_read(Path(path)))

flyloft/config/models.py

@@ -0,0 +1,68 @@
+from pathlib import Path
+from typing import Literal
+
+from pydantic import BaseModel
+
+
+class ResourceProfile(BaseModel):
+    cpus: int = 1
+    mem: str = "1G"
+    time: str = "00:10:00"
+    gpus: int = 0
+
+
+class ProjectConfig(BaseModel):
+    name: str
+    repo_url: str
+    base_branch: str
+    brief_path: str
+    cluster: str
+    run_command: str
+    results_path: str
+    resources: ResourceProfile = ResourceProfile()
+    environment_setup: str = ""
+
+
+class ClusterConfig(BaseModel):
+    name: str
+    host: str
+    user: str
+    remote_work_dir: str
+    account: str | None = None
+    partition: str | None = None
+    password_env: str | None = None
+    container_image: str | None = None
+
+
+class HarnessConfig(BaseModel):
+    name: str
+    kind: Literal["claude-code", "opencode", "scripted"]
+    model: str | None = None
+    base_url_env: str | None = None
+    api_key_env: str | None = None
+    max_turns: int = 30
+    script: str | None = None
+
+
+class FlyloftConfig(BaseModel):
+    database_url: str
+    data_dir: Path
+    tick_interval: float = 15.0
+    max_concurrent_turns: int = 2
+    max_outstanding_jobs: int = 4
+
+
+class RoleSelection(BaseModel):
+    harness: str
+    model: str | None = None
+
+
+class CampaignFile(BaseModel):
+    project: str
+    question: str
+    criterion: str
+    baseline: str
+    constraints: str = ""
+    submission_cap: int
+    roles: dict[str, RoleSelection]
+    cluster: str | None = None

flyloft/publication/body.py

@@ -0,0 +1,28 @@
+from pathlib import Path
+
+from jinja2 import Environment, FileSystemLoader
+
+from flyloft.record.models import Campaign
+from flyloft.record.views import campaign_views
+
+_env = Environment(
+    loader=FileSystemLoader(Path(__file__).parent), autoescape=False, trim_blocks=True, lstrip_blocks=True
+)
+
+
+def pr_title(campaign: Campaign) -> str:
+    return f"flyloft: {campaign.question[:70]}"
+
+
+def render_pr_body(conn, campaign: Campaign) -> str:
+    v = campaign_views(conn, campaign.id)
+    baseline = next((r for r in v.runs if r.purpose == "baseline" and r.state == "finished"), None)
+    return _env.get_template("pr_body.md.j2").render(
+        campaign=campaign,
+        baseline=baseline,
+        runs=v.runs,
+        hypotheses=v.hypotheses,
+        findings=v.findings,
+        critiques=v.critiques,
+        assessments=v.assessments,
+    )

flyloft/publication/gh.py

@@ -0,0 +1,48 @@
+import re
+import subprocess
+
+
+def owner_repo(repo_url: str) -> str:
+    m = re.search(r"github\.com[:/]([^/]+)/([^/]+?)(?:\.git)?/?$", repo_url)
+    if not m:
+        raise ValueError(f"not a GitHub repository URL: {repo_url}")
+    return f"{m.group(1)}/{m.group(2)}"
+
+
+class GhPublisher:
+    def publish(
+        self, *, repo_url: str, base_branch: str, branch: str, title: str, body: str, pr_url: str | None
+    ) -> str:
+        repo = owner_repo(repo_url)
+        if pr_url:
+            subprocess.run(
+                ["gh", "pr", "edit", pr_url, "--repo", repo, "--title", title, "--body-file", "-"],
+                input=body,
+                text=True,
+                check=True,
+                capture_output=True,
+            )
+            return pr_url
+        cp = subprocess.run(
+            [
+                "gh",
+                "pr",
+                "create",
+                "--repo",
+                repo,
+                "--draft",
+                "--base",
+                base_branch,
+                "--head",
+                branch,
+                "--title",
+                title,
+                "--body-file",
+                "-",
+            ],
+            input=body,
+            text=True,
+            check=True,
+            capture_output=True,
+        )
+        return cp.stdout.strip().splitlines()[-1]

flyloft/publication/pr_body.md.j2

@@ -0,0 +1,59 @@
+flyloft campaign `{{ campaign.id }}` on branch `{{ campaign.branch }}`. Campaign state: {{ campaign.state }}.
+
+**Question.** {{ campaign.question }}
+
+**Success criterion.** {{ campaign.criterion }}
+
+**Baseline.** {{ campaign.baseline }}
+{% if baseline %}
+Baseline run `{{ baseline.id }}` at commit `{{ baseline.commit_sha }}`: {{ baseline.metrics }}
+{% endif %}
+
+## Findings
+
+Accepted by the critic.
+{% for f in findings if f.state == "accepted" %}
+
+### {{ f.claim }}
+
+- Hypothesis `{{ f.hypothesis_id }}`
+- Baseline run `{{ f.baseline_run_id }}`: {{ f.baseline_metrics }}
+- Candidate runs: {% for rid in f.candidate_run_ids %}`{{ rid }}` {% endfor %}: {{ f.candidate_metrics }}
+- Commits: {% for r in runs if (r.id | string) in f.candidate_run_ids %}`{{ r.commit_sha }}` {% endfor %}
+- Limitations: {{ f.limitations }}
+{% for c in critiques if c.finding_id == (f.id | string) %}
+- Critic: {{ c.reasoning }}
+{% endfor %}
+{% else %}
+
+None yet.
+{% endfor %}
+
+## Findings under review or superseded
+{% for f in findings if f.state in ["candidate", "blocked", "superseded"] %}
+- [{{ f.state }}] {{ f.claim }} (hypothesis `{{ f.hypothesis_id }}`){% for c in critiques if c.finding_id == (f.id | string) %} Critic: {{ c.reasoning }}{% if c.demands %} Demands: {{ c.demands }}{% endif %}{% endfor %}
+{% else %}
+None.
+{% endfor %}
+
+## Hypotheses not supported or abandoned
+{% for h in hypotheses if h.state in ["refuted", "inconclusive", "unexecuted", "abandoned"] %}
+- [{{ h.state }}] {{ h.title }}: {{ h.change }}.{% for a in assessments if (a.hypothesis_id | string) == (h.id | string) %} {{ a.reasoning }}{% endfor %}{% for f in findings if f.hypothesis_id == (h.id | string) and f.state == "withdrawn" %} Withdrawn finding: {{ f.claim }}{% endfor %}
+{% else %}
+None.
+{% endfor %}
+
+## Open
+{% for h in hypotheses if h.state in ["proposed", "testing"] %}
+- [{{ h.state }}] {{ h.title }}: {{ h.change }}
+{% else %}
+None.
+{% endfor %}
+
+## Runs
+
+| Run | Purpose | Commit | State | Metrics |
+|---|---|---|---|---|
+{% for r in runs %}
+| `{{ r.id }}` | {{ r.purpose }} | `{{ r.commit_sha }}` | {{ r.state }} | {{ r.metrics }} |
+{% endfor %}

flyloft/publication/publisher.py

@@ -0,0 +1,7 @@
+from typing import Protocol
+
+
+class Publisher(Protocol):
+    def publish(
+        self, *, repo_url: str, base_branch: str, branch: str, title: str, body: str, pr_url: str | None
+    ) -> str: ...