discover-artifacts

name	discover-artifacts
version	1.0.0
description	Discover and read artifacts published by other agents
uses	[]
requires	{"tools":[],"env":[]}
security	{"risk_level":"read","capabilities":["artifact:read"],"requires_approval":false}

Discover Artifacts

Find and read work products published by other agents. Artifacts are the primary way agents share results — reviews, analyses, reports, and decisions.

Execution Model

This is a builtin skill (handler: type: builtin). When list_artifacts is invoked, the executor scans artifact directories (.os/memory/{type}/) and returns metadata and paths. The tool schema in tool.yaml defines the external contract; the executor handles the directory listing and frontmatter parsing directly.

When to Use

• Finding work products from other agents (reviews, analyses, reports)
• Checking if someone has already analyzed a topic
• Getting the latest review or report on a subject
• Reacting to new artifacts published on the bus
• Building on previous work rather than starting from scratch

Methodology

1. Determine What You Need

Before searching, clarify:

• Artifact type: What kind of artifact? (review, analysis, report, summary, decision)
• Subject: What topic or entity? (e.g., PR #23, module X)
• Recency: Do you need the latest, or historical?

2. Search by Type

Artifacts are organized by type in directories:

Type	Location	Contains
review	.os/memory/review/	Code review feedback, PR comments
analysis	.os/memory/analysis/	Investigation results, research findings
report	.os/memory/report/	Status reports, progress summaries
summary	.os/memory/summary/	Condensed information, digests
decision	.os/memory/decision/	Architectural decisions, ADRs

3. Read Artifact Metadata

Each artifact has YAML frontmatter with:

• type — artifact type
• subject — what it's about
• created_by — which agent created it
• created_at — when it was created
• related_to — related artifacts (optional)

4. Use the Artifact

After reading, integrate the artifact into your work:

• Don't duplicate work that's already been done
• Reference the artifact in your own output
• Note if the artifact is stale (check created_at)

Output Format

## Artifact Discovery

### Search
- Type filter: [type or all]
- Creator filter: [agent or all]
- Results found: [count]

### Artifacts
| # | Type | Subject | Creator | Created | Path |
|---|------|---------|---------|---------|------|
| 1 | [type] | [subject] | [agent] | [date] | [path] |

### Selected
- Path: [full path]
- Summary: [first few lines of content]

Quality Criteria

• Search is scoped appropriately (not too broad, not too narrow)
• Artifact metadata is parsed correctly from frontmatter
• Results are sorted by relevance or recency
• Stale artifacts are identified (check creation date)
• Artifacts are read and used, not just discovered
• Previous work is referenced rather than duplicated

Common Mistakes

• Not checking for existing artifacts: Starting work from scratch when someone already did the analysis. Always check for artifacts before beginning.
• Ignoring artifact age: Using a week-old review summary when the code has changed since. Check created_at and compare to recent commits.
• Wrong directory: Looking in .os/memory/review/ for a decision record. Use the type taxonomy to find the right directory.
• Duplicating work: Redoing an analysis that another agent already completed and published. Discover first, then build on it.
• Not parsing frontmatter: Skipping the YAML frontmatter and missing important metadata (creator, date, related artifacts).

Completion

Artifact discovery is complete when:

• The correct artifact type and scope are identified
• Available artifacts are found and listed
• Relevant artifacts are read and their content understood
• Artifact metadata (age, creator, relevance) is assessed