agentsclimarketplace

Document spring repo

Skill huntyyyyyy/spring-boot-doc-agent/adapters/claude/skills/document-spring-repo

Scan a Spring Boot repository, ask the user clarifying questions about what static analysis can't determine (write ownership, external consumers, known limitations, intent behind unsecured endpoints), then generate a fixed set of fourteen markdown docs — readme, architecture, integrations, authorization, database, operations, observability, troubleshooting, configuration, change_impact, glossary, local_development, testing, known_limitations. Use whenever the user asks to document a Spring Boot repo, generate onboarding docs for a Java service, map out a legacy Spring codebase, or produce architecture/database/security documentation for a Spring Boot project. This is heavier than the generic document-repo pipeline — use this one specifically for Spring Boot/Spring Data/Spring Security codebases where the fourteen-file taxonomy applies.From its SKILL.md

Install
npx -y skills add huntyyyyyy/spring-boot-doc-agent --skill document-spring-repo

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

6.9 KB, as published. Nobody here has run it

Document Spring Repo

Five conceptual stages. Stage graph SoT: doc_engine.pipeline.stages.build_stage_specs() (deterministic argv + generative choreography metadata via generative_choreography()). This skill owns live generative execution only: Task fan-out to the agents named in that SoT, the interview when requires_human_interview is set, and post-run gates via the CLI.

Prerequisites

  1. doc-engine on PATH (pip install -e . from the product repo). Confirm with doc-engine --help.
  2. Read ${CLAUDE_PLUGIN_ROOT}/CONSTRAINTS.md once (plugin-local stub; points at full monorepo CONSTRAINTS when you have the checkout).
  3. Read ${CLAUDE_PLUGIN_ROOT}/skills/document-spring-repo/references/doc-taxonomy.md before Stage 4.

Do not invoke deterministic tools through the Claude plugin install tree — marketplace installs have no scripts/ directory under the plugin root. Use doc-engine pipeline … only.

Orchestrator entry points (deterministic)

GoalCommand
Stage 0 (then continue with this skill)doc-engine pipeline run <repo_path> --compliance-profile deterministic_only --out-dir <run_dir> --docs-in-target-repo
Stop after a named stagedoc-engine pipeline run <repo_path> --until <stage> --out-dir <run_dir>
Fast signal-scan smokedoc-engine pipeline run <repo_path> --compliance-profile scan_only --out-dir <run_dir>
Full mock E2E (CI / wiring check)doc-engine pipeline run <repo_path> --out-dir <run_dir>
Gates after live generative stagesdoc-engine pipeline gates --out-dir <run_dir> --target-repo <repo_path> --docs-dir <repo_path>/docs
Certification checkdoc-engine certification verify <run_dir>/certification.json

Stage names for --until and generative agent bindings come from build_stage_specs() / generative_choreography() — do not maintain a second stage list here.

Every pipeline run writes certification.json under --out-dir. certified: true with generative_executor: mock means structural wiring passed — not human-quality docs. Live Claude runs must complete generative stages below and then pipeline gates.

Target repos may set .doc-engine.yml:

compliance_profile: certified

Data contracts

Artifacts cross stage boundaries. Shapes are enforced by Pydantic models in doc_engine.pipeline.artifacts. The orchestrator validates at Stage 0 boundaries; after live generative stages, run doc-engine pipeline gates.

ArtifactProducerConsumers
spring_signals.jsonStage 0 (CLI)generative stages
facts.jsonlStage 0 (CLI, dual-emit)gap_analysis_interview / doc_writer (contested identity)
groups.jsonStage 0 (CLI)file_summarize / architect
cross_group_edges.jsonStage 0 (CLI)file_summarize
summaries.jsonfile_summarize agentslater generative stages
interview_answers.jsongap_analysis_interviewdoc_writer

Work in --out-dir (and --docs-in-target-repo for docs/). Manifest and signals land beside each other in the run directory.

Agents — bind via generative_choreography() (file-summarizer, doc-writer, gap-analyzer, architect-segment, architect-merge, software-architect-and-testing) under ${CLAUDE_PLUGIN_ROOT}/agents/, dispatched by name via Task.

Stage 0 — Deterministic evidence (CLI only)

doc-engine pipeline run <repo_path> \
  --compliance-profile deterministic_only \
  --out-dir <run_dir> \
  --docs-in-target-repo

Optional drift pre-check: python -m doc_engine.tools.spring_drift_check. Prefer re-running Stage 0 when unsure. Boundary validation uses validate_artifacts.py / python -m doc_engine.tools.validate_artifacts.

Also collect TODO/FIXME/XXX/HACK candidates for known_limitations.md via structural search (ast-grep on Java — comments are AST nodes) or from file-summarizer evidence; do not use text grep/rg (denied for agents). Keep hits as candidates, not facts — see doc-taxonomy.md.

Read spring_signals.json, facts.jsonl, groups.json, and cross_group_edges.json from <run_dir> before generative work. Prefer doc-engine query … for bounded lookups (see ${CLAUDE_PLUGIN_ROOT}/SEARCH.md).

Generative stages (live adapter)

Follow generative_choreography() order. Summary (must match the SoT — if prose and SoT disagree, SoT wins):

  1. file_summarize — agent file-summarizer per group → concatenate summaries_group_*.json into summaries.json.
  2. architect — architect-segment per group, then architect-merge.
  3. gap_analysis_interview — gap-analyzer + software-architect-and-testing, then human interview in this thread (requires_human_interview) → interview_answers.json.
  4. doc_writer — doc-writer × fourteen taxonomy files under docs/.

Concatenate per-group summaries:

python3 -c "import json,glob,os; d=os.environ['RUN_DIR']; json.dump([o for f in sorted(glob.glob(os.path.join(d,'summaries_group_*.json'))) for o in json.load(open(f))], open(os.path.join(d,'summaries.json'),'w'), indent=1)"

Interview answer shape:

[
  {"id": "integrations.who-calls-us", "question": "...", "status": "answered", "answer": "...", "date": "2026-07-24"},
  {"id": "known_limitations.retry-policy", "question": "...", "status": "skipped", "answer": null, "date": "2026-07-24"}
]

Never overwrite a root README.md — use docs/readme.md when a root README exists.

Finish (gates + certification)

doc-engine pipeline gates \
  --out-dir <run_dir> \
  --target-repo <repo_path> \
  --docs-dir <repo_path>/docs

doc-engine certification verify <run_dir>/certification.json

Do not tell the user the run succeeded while gates fail.

What this deliberately does not do

  • No plugin-local deterministic tool tree — tools live in the doc-engine package.
  • No duplicating build_stage_specs() as bash one-liners in this file.
  • No automatic cross-repo discovery beyond the interview.
  • Regenerating docs remains a deliberate re-run of Stage 0 + generative stages.

Keep looking

Skills are one crate of 608,666. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.