Skip to content

feat(demo): add an installed skill-library demo (--kind skill_library) #881

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Birfy wants to merge 1 commit into
base: main
Choose a base branch
from
+454 −184
Open

feat(demo): add an installed skill-library demo (--kind skill_library) #881

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -105,7 +105,7 @@ print(result.outcomes()) # why it went that way
On 40 HotpotQA items (12 held out) this took held-out exact match from **0.167
to 0.583**, learning *"Respond with only the requested answer, omitting any extra
explanation or restatement."* That evolves an **instruction** (`SingleSlot`), not
an Agent Skill folder — for a library run `python -m examples.skill_library_evolution --dry-run`.
an Agent Skill folder — for a library run `agentdescent demo --kind skill_library`.
Pass `agent=LLMAgent(claude_code())` (or any
[agent or LLM](https://thinkflowlab.github.io/agentdescent/agents/)) instead of
`run`/`propose` to evolve against a tool-using CLI agent.
Expand Down Expand Up @@ -203,7 +203,7 @@ for you to check first.
`agent_dir`, `agent_code`, `policy_slot` — or **`plugin`**: the host plugins
themselves are evolvable, each rollout in an isolated copy of the host with
hooks and permissions frozen and a recursion guard. Flagship library demo:
`python -m examples.skill_library_evolution --dry-run`.
`agentdescent demo --kind skill_library`.
[The plugin, step by step →](https://thinkflowlab.github.io/agentdescent/plugin-quickstart/)
· [set it up and test it →](https://thinkflowlab.github.io/agentdescent/testing-guide/)

Expand Down
60 changes: 51 additions & 9 deletions agentdescent/cli.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -681,10 +681,17 @@ def cmd_demo(a: argparse.Namespace) -> int:
f"{list(demo_mod.OFFLINE_ALGORITHMS)}. Run it for real with "
f"`agentdescent algorithms run {algo} <path> <cases.jsonl>` "
f"(`agentdescent algorithms show {algo}` for its parameters)")
try:
demo_mod.resolve_demo_kind(algo, getattr(a, "kind", None))
except ValueError as e:
raise SpecError(str(e)) from None
root = os.path.abspath(os.path.expanduser(a.dir)) if a.dir else tempfile.mkdtemp(
prefix="agentdescent-demo-")
os.makedirs(root, exist_ok=True)
spec_dict = demo_mod.build(root, algorithm=algo)
try:
spec_dict = demo_mod.build(root, algorithm=algo, kind=getattr(a, "kind", None))
except ValueError as e:
raise SpecError(str(e)) from None
kind = spec_dict["kind"]

as_json = getattr(a, "json", False)
Expand All @@ -695,6 +702,15 @@ def cmd_demo(a: argparse.Namespace) -> int:
print(f" prompt.txt -> {open(prompt_path).read().strip()!r} "
f"(wrong: `id` is a row number)")
print(" cases.jsonl -> 12 prompts with embedded CSVs and known totals")
elif kind == "skill_library":
csv_rules = os.path.join(root, "csv-total", "references", "rules.md")
date_rules = os.path.join(root, "date-parse", "references", "rules.md")
print(f"A two-skill library (csv-total + date-parse), in {root}")
print(f" csv-total/references/rules.md -> "
f"{open(csv_rules).read().strip()!r} (wrong: `id` is a row number)")
print(f" date-parse/references/rules.md -> "
f"{open(date_rules).read().strip()!r} (wrong: date tasks ask for year)")
print(" cases.jsonl -> 12 tasks (CSV totals + date years)")
else:
skill = os.path.join(root, demo_mod.DEMO_SKILL)
rules = os.path.join(skill, "references", "rules.md")
Expand All @@ -703,8 +719,9 @@ def cmd_demo(a: argparse.Namespace) -> int:
f"{open(rules).read().strip()!r} (wrong: `id` is a row number)")
print(" cases.jsonl -> 12 CSVs with known totals")
print(" the agent is a local program, so this costs nothing and needs no key")
print(" note: this is a smoke trajectory; for multi-round library edits see")
print(" python -m examples.skill_library_evolution (#847)")
if kind != "skill_library":
print(" note: this is a smoke trajectory; for multi-round library edits see")
print(" agentdescent demo --kind skill_library")
if algo:
print(f" algorithm: {algo} (`agentdescent algorithms show {algo}`)")
print()
Expand Down Expand Up @@ -737,6 +754,14 @@ def cmd_demo(a: argparse.Namespace) -> int:
"agentdescent doctor",
"agentdescent plan spec.json",
]
elif kind == "skill_library":
next_steps = [
f"agentdescent show {rd.run_id}",
f"agentdescent apply {rd.run_id} --dry-run",
"agentdescent init <your-skill-library> --kind skill_library --data <your-cases.jsonl> --out spec.json",
"agentdescent doctor",
"agentdescent plan spec.json",
]
else:
next_steps = [
f"agentdescent show {rd.run_id}",
Expand Down Expand Up @@ -799,6 +824,19 @@ def cmd_demo(a: argparse.Namespace) -> int:

agentdescent init <your-prompt.txt> --kind text --data <your-cases.jsonl> --out spec.json
agentdescent doctor
agentdescent plan spec.json""")
elif kind == "skill_library":
print(f"""
The library on disk is untouched -- an evolved artifact only lands when you say so:

agentdescent show {rd.run_id} the diff
agentdescent apply {rd.run_id} --dry-run what it would write

Your own run is the same spec with three things changed -- your library, your
cases.jsonl, and a real agent instead of the offline one:

agentdescent init <your-skill-library> --kind skill_library --data <your-cases.jsonl> --out spec.json
agentdescent doctor # what is missing before a real run
agentdescent plan spec.json""")
else:
print(f"""
Expand Down Expand Up @@ -1239,10 +1277,10 @@ def write_starter_spec(
if spec["kind"] == "skill_library":
# #864 / #846: demo is skill_dir smoke; name the library card path.
report.append(("next",
"(`agentdescent demo --algorithm archive_selection` "
"is an offline skill_dir smoke; for this library "
"Path A card use --algorithm archive_selection — "
"see docs/quickstart-skill.md)"))
"(`agentdescent demo --kind skill_library` is "
"the offline library demo; `demo --algorithm "
"archive_selection` is a one-skill smoke of the "
"same card — see docs/quickstart-skill.md)"))
else:
report.append(("next", "(`agentdescent demo` runs a complete example, "
"offline, to copy from)"))
Expand Down Expand Up @@ -4608,10 +4646,14 @@ def build_parser() -> argparse.ArgumentParser:

s = sub.add_parser("demo", help="build a complete example and run it, offline")
s.add_argument("--dir", help="where to build it (default: a temp directory)")
s.add_argument("--kind", choices=("skill_dir", "text", "skill_library"),
help="demo fixture (default: skill_dir, or text when "
"--algorithm is a text-only card)")
s.add_argument("--algorithm", metavar="NAME",
help="run the demo with this named algorithm offline "
"(skill_dir: archive_selection; text: gepa, dgm, ace, "
"skillopt, rrsi; evoskill is refused — needs a real model). "
"(skill_dir / skill_library: archive_selection; text: "
"gepa, dgm, ace, skillopt, rrsi; evoskill is refused — "
"needs a real model). "
"Currently: " + ", ".join(_demo_offline_algorithms()))
s.set_defaults(fn=cmd_demo)

Expand Down
Loading
Loading