workflow_skill_creator

Name: workflow_skill_creator
Author: ai4protein

244stars35forks

How to install workflow_skill_creator?

ai4protein/venusfactory2/workflow_skill_creator

$npx -y skills add ai4protein/VenusFactory2 --skill workflow_skill_creator

Installs into the current project.

Summary

Distills a completed user workflow or interaction into a reusable VenusFactory agent skill. Use when the user says "make this a skill", "create a skill from what we just did", "package this workflow" or similar. Adapts the workflow into the VenusFactory tools wiring + SKILL.md pa

Files · 1

View on GitHub

SKILL.md

1	# Workflow-to-Skill Distiller
2
3	Turns a completed VenusFactory workflow into a reusable agent skill. Extracts patterns from an interaction that already happened and packages them following the project's conventions.
4
5	> [!CAUTION] You MUST complete Phase 1 (Brainstorming) before writing any code or SKILL.md content. Skipping brainstorming produces skills that are either too rigid or too vague.
6
7	## Phase 1: Brainstorming (MANDATORY)
8
9	Have an iterative back-and-forth conversation with the user. Do NOT ask all questions at once. Pick 2-3 relevant questions per round, refine your understanding, and ask follow-ups.
10
11	### Round 1: Understand the Workflow
12
13	Start by summarizing what you observed, then ask:
14
15	1. "Here's my understanding of the workflow: [summary]. Is this accurate?"
16	2. "What are the expected inputs and outputs?"
17	3. "How often will this run? One-off, recurring, or part of a larger pipeline?"
18
19	### Round 2: Flexibility and Error Handling
20
21	For each step:
22
23	1. "If [step X] fails (API down, no results), should the agent (a) ask for guidance, (b) try alternatives automatically, or (c) fail loudly?"
24	2. "Are there steps where the exact method matters (must use a specific database / model), vs. steps where any reasonable approach is fine?"
25
26	### Round 3: Reuse Existing VenusFactory Tools
27
28	Before asking these questions, check `src/tools/database/`, `src/tools/visualize/`, `src/tools/predict/`, `src/tools/mutation/`, `src/tools/search/`, `src/tools/train/`, `src/tools/file/`, `src/tools/bioinfo/` for overlap. If an existing tool covers a step, the new skill MUST reference it — do not reimplement.
29
30	1. "I noticed the workflow uses [tool X, tool Y] that already exist. The new skill will reference these. Anything else to incorporate?"
31	2. "Are there rate limits for new external APIs that aren't covered?"
32	3. "Any reference docs (API specs, papers, datasets) I should include under `references/`?"
33
34	### Round 4: Scope, Code, and Naming
35
36	1. "Should the skill cover [X, Y, Z] from the workflow, or include/exclude anything?"
37	2. Determine whether the skill needs new code:
38	- Needs new code if any step calls an external API not yet wrapped, processes files, or computes results not already in `src/tools/`. → Follow the 6-file wiring recipe below.
39	- Instruction-only if every step orchestrates existing VenusFactory tools. → Write SKILL.md only, no Python.
40	3. Propose a name: `{verb}_{noun}_database` for new DB wrappers, `{noun}` for analysis/visualization, matching existing folders (`alphafold_database`, `pymol`, `clustalo_msa`, etc.).
41
42	### Round 5: Validation (Optional)
43
44	1. "Sample query + expected answer I can use to verify? Optional but helpful."
45
46	### Brainstorming Completion Criteria
47
48	Move to Phase 2 only when you can answer:
49
50	- [ ] Purpose and scope
51	- [ ] Inputs and outputs
52	- [ ] Strict vs flexible steps
53	- [ ] Which existing VenusFactory tools are reused
54	- [ ] What new scripts (if any) are needed
55	- [ ] Rate limits
56	- [ ] Error handling strategy
57	- [ ] Code needed (→ 6-file wiring) or instruction-only (→ SKILL.md only)
58	- [ ] Sample query/answer
59
60	## Phase 2: Skill Design
61
62	Produce a design document (markdown plan) and present for approval:
63
64	1. Skill name and frontmatter (see Rule 6).
65	2. Directory structure showing all planned files.
66	3. Existing VenusFactory tools referenced with rationale.
67	4. New code files with proposed function signatures.
68	5. Rate limiting strategy for any new external API.
69	6. Error handling strategy per step.
70
71	Wait for explicit user approval before Phase 3.
72
73	## Phase 3: Implementation
74
75	### Guiding Principles
76
77	- Match the project's existing Python style — use the conda env at `~/miniconda3/envs/venus/bin/python` (see `environment.yaml`); do NOT introduce `uv` or per-script `# /// script` headers.
78	- Prefer `requests` + `urllib3.util.retry.Retry` for HTTP (this is the established pattern, e.g. `src/tools/database/alphafold/alphafold_structure.py`).
79	- All public download/query functions return a JSON string with the rich envelope:
80	- Success: `{"status": "success", "file_info": {...} \| "content": "...", "content_preview": str, "biological_metadata": {...}, "execution_context": {...}}`
81	- Error: `{"status": "error", "error": {"type", "message", "suggestion"}, "file_info": null}`
82	- Use `to_client_file_path()` from `src/tools/path_sanitizer.py` for any `file_path` in responses.
83	- Document rate limits in comment

People who liked this also viewed

6 picks

Type

workflow_skill_creator

244stars35forks

View on GitHub

How to install workflow_skill_creator?

ai4protein/venusfactory2/workflow_skill_creator

$npx -y skills add ai4protein/VenusFactory2 --skill workflow_skill_creator

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Workflow-to-Skill Distiller
2
3	Turns a completed VenusFactory workflow into a reusable agent skill. Extracts patterns from an interaction that already happened and packages them following the project's conventions.
4
5	> [!CAUTION] You MUST complete Phase 1 (Brainstorming) before writing any code or SKILL.md content. Skipping brainstorming produces skills that are either too rigid or too vague.
6
7	## Phase 1: Brainstorming (MANDATORY)
8
9	Have an iterative back-and-forth conversation with the user. Do NOT ask all questions at once. Pick 2-3 relevant questions per round, refine your understanding, and ask follow-ups.
10
11	### Round 1: Understand the Workflow
12
13	Start by summarizing what you observed, then ask:
14
15	1. "Here's my understanding of the workflow: [summary]. Is this accurate?"
16	2. "What are the expected inputs and outputs?"
17	3. "How often will this run? One-off, recurring, or part of a larger pipeline?"
18
19	### Round 2: Flexibility and Error Handling
20
21	For each step:
22
23	1. "If [step X] fails (API down, no results), should the agent (a) ask for guidance, (b) try alternatives automatically, or (c) fail loudly?"
24	2. "Are there steps where the exact method matters (must use a specific database / model), vs. steps where any reasonable approach is fine?"
25
26	### Round 3: Reuse Existing VenusFactory Tools
27
28	Before asking these questions, check `src/tools/database/`, `src/tools/visualize/`, `src/tools/predict/`, `src/tools/mutation/`, `src/tools/search/`, `src/tools/train/`, `src/tools/file/`, `src/tools/bioinfo/` for overlap. If an existing tool covers a step, the new skill MUST reference it — do not reimplement.
29
30	1. "I noticed the workflow uses [tool X, tool Y] that already exist. The new skill will reference these. Anything else to incorporate?"
31	2. "Are there rate limits for new external APIs that aren't covered?"
32	3. "Any reference docs (API specs, papers, datasets) I should include under `references/`?"
33
34	### Round 4: Scope, Code, and Naming
35
36	1. "Should the skill cover [X, Y, Z] from the workflow, or include/exclude anything?"
37	2. Determine whether the skill needs new code:
38	- Needs new code if any step calls an external API not yet wrapped, processes files, or computes results not already in `src/tools/`. → Follow the 6-file wiring recipe below.
39	- Instruction-only if every step orchestrates existing VenusFactory tools. → Write SKILL.md only, no Python.
40	3. Propose a name: `{verb}_{noun}_database` for new DB wrappers, `{noun}` for analysis/visualization, matching existing folders (`alphafold_database`, `pymol`, `clustalo_msa`, etc.).
41
42	### Round 5: Validation (Optional)
43
44	1. "Sample query + expected answer I can use to verify? Optional but helpful."
45
46	### Brainstorming Completion Criteria
47
48	Move to Phase 2 only when you can answer:
49
50	- [ ] Purpose and scope
51	- [ ] Inputs and outputs
52	- [ ] Strict vs flexible steps
53	- [ ] Which existing VenusFactory tools are reused
54	- [ ] What new scripts (if any) are needed
55	- [ ] Rate limits
56	- [ ] Error handling strategy
57	- [ ] Code needed (→ 6-file wiring) or instruction-only (→ SKILL.md only)
58	- [ ] Sample query/answer
59
60	## Phase 2: Skill Design
61
62	Produce a design document (markdown plan) and present for approval:
63
64	1. Skill name and frontmatter (see Rule 6).
65	2. Directory structure showing all planned files.
66	3. Existing VenusFactory tools referenced with rationale.
67	4. New code files with proposed function signatures.
68	5. Rate limiting strategy for any new external API.
69	6. Error handling strategy per step.
70
71	Wait for explicit user approval before Phase 3.
72
73	## Phase 3: Implementation
74
75	### Guiding Principles
76
77	- Match the project's existing Python style — use the conda env at `~/miniconda3/envs/venus/bin/python` (see `environment.yaml`); do NOT introduce `uv` or per-script `# /// script` headers.
78	- Prefer `requests` + `urllib3.util.retry.Retry` for HTTP (this is the established pattern, e.g. `src/tools/database/alphafold/alphafold_structure.py`).
79	- All public download/query functions return a JSON string with the rich envelope:
80	- Success: `{"status": "success", "file_info": {...} \| "content": "...", "content_preview": str, "biological_metadata": {...}, "execution_context": {...}}`
81	- Error: `{"status": "error", "error": {"type", "message", "suggestion"}, "file_info": null}`
82	- Use `to_client_file_path()` from `src/tools/path_sanitizer.py` for any `file_path` in responses.
83	- Document rate limits in comment

People who liked this also viewed

6 picks

Type