Module map

A reference to every module in src/skillgen/, taken from the source. Each public function has a Google-style docstring in the code; this page is the map, the code is the detail.

Top level

Module	Role	Key names
__init__.py	Package constants	TOOL_NAME = "skillgendev", __version__
cli.py	Command line; the only module that prints	main, build_parser, HANDLERS, emit, Reply, EXIT_OK/FAILED/ERROR/NEEDS, command_<name>
pipeline.py	Per-route orchestration	discover_routes, prepare_route, analyze_route, finish_route, write_skill, fingerprint_request, measure_page_model, write_index, ask_claude_for_routes
handoff.py	The next step a stopped run hands over	NextStep, build_next_step, format_command
workspace.py	.skillgendev/ paths and atomic file access	Workspace, find_workspace_root, write_text_atomically, GITIGNORE_LINES
config.py	config.json	Settings, SliceSettings, Environment, AnalyzerKind, check_app_url
assets.py	Shipped data in a wheel or checkout	find_data_folder, read_fill_engine, read_index_template
export.py	Exports, manifests, plugin layouts	export_skills, export_chatgpt_plugin, check_export_safety, build_manifest, PLUGIN_AUTHOR
logs.py	Run logs	start_logging, list_logs, KEPT_LOGS = 30
errors.py	Exceptions and needs	SkillgenError and subclasses, Need, ActionNeededError

analyze/

Module	Role	Key names
prompt.py	The prompt Claude Code receives for one route, and for listing routes	PROMPT_VERSION = 3, AnalysisRequest, build_analysis_prompt, build_route_listing_prompt
runner.py	Headless Claude Code, a file Claude wrote, or a replay	ClaudeCodeAnalyzer, FileAnalyzer, ReplayAnalyzer, build_claude_command, READ_ONLY_TOOLS, is_limit_reached

graph/

Module	Role	Key names
codegraph.py	File-level code graph built with Graphify, with a staleness snapshot	CodeGraph, build_code_graph, find_graph_staleness, find_route_nodes
connections.py	Graphify's own explain and path, through its command line	GraphifyGraph, explain_node, find_path
hubs.py	Ranks files that probably declare routes (pure)	rank_router_candidates, RouterHint
slice.py	Page slices and change impact (pure)	slice_page, find_affected_files, find_stale_skills, resolve_component

routes/

Module	Role	Key names
hintpacks.py	Loads YAML hint packs	load_hint_packs, FolderPack, RouterPack
folders.py	Routes from folder-routing packs, monorepo roots	find_folder_routes, find_project_roots
astgrep.py	Routes declared in code, through ast-grep	find_declared_routes, run_ast_grep, is_in_skipped_folder
merge.py	Merges candidates into routes.json, keeping developer edits	merge_routes, rename_route, find_pattern_files, check_skill_name
naming.py	Skill names and descriptions under the Agent Skills rules	build_skill_name, is_valid_name, build_description, MAX_NAME_LENGTH = 64
paths.py	One canonical route shape	normalize_route, match_route, check_route_text, is_route_wanted

model/

Module	Role	Key names
page.py	The page model	PageModel, FieldSpec, Strategy, ValueType, Condition, View, Layout, SubmitSpec
facts.py	Sources and evidence	Source, Provenance, Evidence
confidence.py	Confidence arithmetic	combine_confidence, cap_conflict, INCLUSION_THRESHOLD = 0.7, CONFLICT_CAP = 0.5
route.py	routes.json and questions.json entries	RouteEntry, RouteTable, Question
serialize.py	page.json to and from the model	page_to_dict, page_from_dict
probe.py	What the probe saw, as plain data	ProbedPage, ProbedField

compile/

Module	Role	Key names
merge.py	Analysis + probe + answers → page model	build_page_model, merge_open_steps, merge_views, is_risky_step
fields.py	Merges one field	assess_field, build_field, collect_questions
assess.py	Decides one fact and applies answers	assess_fact, apply_answer
layout.py	Attaches measurements as a layout	build_layout, carry_layout
unopened.py	What the live page had that the model missed	find_missed_on_page

generate/

Module	Role	Key names
skill_md.py	Renders SKILL.md and references/fill.md	render_skill_md, render_fill_reference, BODY_TOKEN_BUDGET = 5000
inline_setter.py	The setter SKILL.md carries	build_inline_setter, render_engine, choose_landmarks, STATUSES, LAYOUT_STATES
fill_script.py	scripts/fill.js	build_fill_script, find_forbidden_syntax, MAX_SCRIPT_BYTES = 48000
fill_engine.js	The shared fallback engine
locators.py	references/locators.md	render_locators
manual_steps.py	references/manual-steps.md	render_manual_steps
layout_reference.py	references/layout.json	render_layout_reference
examples.py	Synthetic example values from constraints	build_example_values
evals.py	evals/evals.json	build_evals
folder.py	All files of one skill, in memory	build_skill_files, fit_skill_text
bundle.py	Many page skills into one skill	build_bundle_files, build_entry_skill_md
user_rules.py	The user block and check: annotations	extract_user_block, parse_check, append_user_rule
index_page.py, index_template.html	skills-index.json and .html	build_index_data, render_index_html

probe/

Module	Role	Key names
runner.py	Playwright: probe, measure, login	probe_route, measure_page_layout, find_xy_need, save_login, CAPTURE_VIEWPORT
extract.js	In-page extraction of fields and clickables
measure.js	Measures a skill's own selectors
classify.py	Raw controls → probed fields (pure)	classify, describe_shape
capture.py	Saves and reads probe captures	probed_page_to_dict
journal.py	What discovery already learnt about a page	Journal, record_state, read_journal

validate/, drift/, schema/

Module	Role	Key names
validate/spec.py	Agent Skills spec checks	check_skill_md, check_skill_files, check_fill_script, MAX_BODY_LINES = 500
validate/model_checks.py	Page model consistency	check_page_model, find_cycle
validate/scan.py	Secret and personal-data scan	scan_files, SAFE_EMAIL_DOMAINS, SAFE_PHONE
drift/check.py	Stale skills after code changes	find_drift, list_changed_files, IMPACT_DEPTH = 2
drift/diff.py	Semantic diff of two page models (pure)	diff_pages, Change
schema/	JSON schemas and one validator	validate_document, find_schema_problems