Skip to content
Agentmart
Submit a tool

prior-art

Find what already settles part of a topic, in the docs sources the project configured, the repository, its beads, and (with the user's permission) other…

Skill by Igosuki

IndexedMIT
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/igosuki-prior-art

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Find what already settles part of a topic, in the docs sources the project configured, the repository, its beads, and (with the user's permission) other connected sources such as wikis, Notion or Drive. Read-only; it reports where each item lives and what it establishes, plus what was searched. Used by sdlc:design and sdlc:split; use it alone to answer "have we done this before?".

Use cases: design, productivity

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/prior-art

Tools it pre-approves 4

From allowed-tools in SKILL.md: the agent may use these without asking you each time.

  • Bash(${CLAUDE_PLUGIN_ROOT}/scripts/settings.sh
  • sources),
  • Bash(bd
  • *)

What's inside

Skill name
prior-art
Description
Find what already settles part of a topic, in the docs sources the project configured, the repository, its beads, and (with the user's permission) other connected sources such as wikis, Notion or Drive. Read-only; it reports where each item lives and what it establishes, plus what was searched. Used by sdlc:design and sdlc:split; use it alone to answer "have we done this before?".
License
MIT
Allowed tools
Bash(${CLAUDE_PLUGIN_ROOT}/scripts/settings.sh, sources),, Bash(bd, *)
Instructions
73 lines in SKILL.md · view on GitHub

Before installing

  • Pre-approves some shell commands

Files (1)

  • SKILL.md

SKILL.md (first lines)

# Prior art

Topic: $ARGUMENTS

Configured sources, in order:
!`${CLAUDE_PLUGIN_ROOT}/scripts/settings.sh sources 2>&1 || true`

Prior art is anything that already settles part of the topic: earlier designs, docs, code, existing beads, or knowledge that lives outside the repository. Find it, report it, and change nothing: no beads, no commits, no files.

`--local` means: the configured sources, the repository and its beads only. Skip step 4.

## 1. Form the queries

From the topic, before searching, write down:
- a **term list**: the domain nouns, plus the repository's own words for them (glossary, README) and one synonym each
- a **sentence**: what the topic does, in one line, for semantic search
- the **symbols and paths** it names, if any

## 2. Find the routes

Each kind of search answers one kind of question and takes one kind of query. Look for tools in your context, then through tool search with `search`, `semantic`, `index`, `graph`, `ast` and `read`. Route by question:

| Question | Route | Query | Fallback when no such tool |
|---|---|---|---|
| Was this designed or discussed before? | full-text search over docs and beads; semantic search if present | the term list; the sentence for semantic | text search under the docs directories; `bd search` |
| Where does the code do this? | semantic or code search | the sentence | text search on the terms |
| What touches this symbol or file? | code graph | the symbol or path | text search on the symbol |
| Every place a pattern occurs? | structural or AST search | a pattern | text search with a regex |
| What is in this large file? | bulk reader | a question | `sed -n` |

Index-backed routes first, in parallel, one query each. Text search only for what they miss or when no such tool is installed: the harness's own search tool when it has one, else `rg`, else `grep`.

## 3. Search, in this order

1. **Configured sources**, in the order listed above. Each is consented by configuration: search it without asking, in a headless session too. `via: path` is a directory, searched with the local routes whether or not it is inside the repository. Any other `via` is a word for tool search; when no tool matches it, the source is unreachable: say so in the `Searched` line and under open questions, and move on. `holds` says what the source can answer; skip it for questions it can't.
2. **The repository**: earlier designs in `docs/design/` and the repository's own convention, such as `docs/adr/`, `specs/`, `rfcs/` or `design/`; `README*` and other docs; the code the topic touches: entry points, data formats, public interfaces, config. Skim; don't read everything.
3. **Beads**, if the repository uses them. Read-only commands only: `bd search <terms>` once per term set, `bd list --type epic`, `bd show <id>` on hits.
4. **Other external sources**, ask first (below): MCP servers such as wikis, Notion, Google Drive, Confluence, issue trackers, chat, mail; skills and commands that fetch knowledge; web search. Find them through tool search with `notion`, `drive`, `wiki`, `confluence`, `jira`, `slack`, `web`. Web search comes last, and only for a question steps 1 to 3 leave open that the web can settle, such as a library's behaviour or a standard; never for the project's own history.

Links given in the topic are always read.

### Ask before using other external sources

- **When you can ask:** use one AskUserQuestion (`multiSelect`) listing the external sources that look relevant to this topic. For each option, say what you would look for there. Offer at most 4; group similar sources if there are more. Consult only the sources the user selects. Then offer, in a second question, to save the selected ones as configured sources: shared ones in `.claude/sdlc.md`, this machine's paths in `.claude/sdlc.local.md`, one `sources:` entry each with `name`, `via`, `where`, `holds`. A saved source is consented for every later search.
- **When you cannot ask** (headless session): use only the external sources the topic names or links to. List the rest under open questions instead of consulting them.

## 4. Follow each hit one hop

A doc's links and its prior-art section. A bead's `metadata.design`, its parent and its last comments. A symbol's callers. Most prior art is found by reference, not by term.

## 5. Stop

Stop when a round of queries adds no new item.

## 6. Report

In the reply, in this form:

```markdown
## Prior art

- `path` | `bead:<id>` | `<source>:<title or url>`: what it establishes.

(or "No prior art.")

Searched: <routes run and the tools used>; sources: <configured sources consulted, and any unreachable one with why>; external: <other sources consulted, or none>.

Open questions: <sources not consulted and why, or none>.
```

Say what each item establishes, not how it relates to new work: whether to reuse, extend, replace or avoid it is the caller's decision.

From Igosuki/agentic-sdlc, MIT.

More from Igosuki/agentic-sdlc

Signals

Stars
0
Forks
0
Open issues
0
Last commit (this folder)
Sep 21, 2026 (1 month ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from Igosuki

Cite this page

prior-art skill. Agent Mart. https://agentmart.dev/skills/igosuki-prior-art (last updated 2026-10-24)