| name | artefact-retrieval |
| description | Search outside the project (in IDE plans, AI agent workspaces, or temporary storage) and retrieve standalone artefacts, research, or validation logs to bring them inside the project repository. |
SKILL_artefact-retrieval
Overview
IDE-agnostic pattern for retrieving valuable artefacts (plans, analyses, validation reports, research notes) that AI coding assistants generate outside the project tree, and importing them into the project's docs/artefacts/ taxonomy for permanent retention and cross-session reuse.
Host Environments
This skill works identically whether the agent is invoked from:
- Claude Code (CLI, IDE extension, or Desktop app)
- Antigravity (Gemini brain workspace)
- Cursor (Plan Mode, Composer, or Agent)
- Any combination — same project, multiple assistants across sessions
The retrieval logic is source-agnostic: each IDE/assistant writes its artefacts to a known location outside the repo, and this skill catalogues those locations so any agent can sweep them into the project artefact tree.
⚠️ Claude Code auto-memory: consult, don't copy
~/.claude/projects/*/memory/ is personal memory, not an artefact store — memory files stay where they are and are never copied into docs/artefacts/.
But memory informs retrieval. Before sweeping IDE locations, skim:
MEMORY.md and any project_*.md entries for the active project — they often name current initiatives, IDEA numbers, modules, stakeholders, and deadlines. Use these to seed PROJECT_KEYWORDS for the discovery sweep (§2) and the unsaved-artefacts audit (§5).
user_*.md / feedback_*.md for signals about which assistant (Cursor / Antigravity / Claude Code) the user has been using recently — narrows where to look first.
The same consult-don't-copy rule applies to ~/.claude/CLAUDE.md and private user config. ~/.gemini/ / ~/.cursor/ config, credentials, and session metadata remain fully off limits — only generated work product (plans, analyses, walkthroughs) from those workspaces is a retrieval candidate.
When to Use
- 🆕 Starting a new project and looking for validated knowledge from prior work
- 🔍 Researching proven patterns before implementing a solution
- ✅ Validating assumptions against documented findings
- 🧹 Periodic sweep for completed plans and analyses that still live outside the repo
- 🔄 Ensuring consistency across agents (Claude Code / Antigravity / Cursor all see the same canonical knowledge)
Artefact Sources (Complete Reference)
| Source | Location | Format |
|---|
| In-repo agent outputs | docs/artefacts/by-agent/ | Markdown |
| In-repo research / validations | docs/artefacts/by-type/ | Markdown |
| In-repo topic cross-refs | docs/artefacts/by-topic/ | Symlinks |
| Cursor plans | ~/.cursor/plans/*.plan.md | YAML frontmatter + Markdown |
| Cursor agent transcripts | ~/.cursor/projects/<project>/agent-transcripts/ | JSONL |
| Antigravity/Gemini artefacts | ~/.gemini/antigravity/brain/<id>/artifacts/ | Markdown |
| Antigravity/Gemini transcripts | ~/.gemini/antigravity/brain/<id>/.system_generated/logs/overview.txt | Text |
| Claude Code transcripts | ~/.claude/projects/<project-slug>/*.jsonl | JSONL (session logs) |
| Temp scratch | /tmp/*.md, /tmp/artefacts/ | Markdown |
Excluded from copying (consult-only or fully off-limits)
- 📖
~/.claude/projects/*/memory/ — personal agent memory; consult project-type entries for keywords and scope, but never copy into docs/artefacts/
- ❌ Any
.env, credentials, session tokens — fully off-limits
- ❌ Raw transcripts — search them for artefact references, but don't copy wholesale
Pattern
1. Pick the source
Figure out which assistant generated the artefact. If unsure, sweep all known locations (§2).
2. Discovery — list and filter
ls -lt ~/.cursor/plans/*.plan.md 2>/dev/null
ls -lt ~/.gemini/antigravity/brain/ 2>/dev/null
for f in ~/.cursor/plans/*.plan.md; do
name=$(grep "^name:" "$f" | head -1 | sed 's/name: //')
echo "$f → $name"
done
KEYWORDS="myproject|IDEA-|ai_service|dashboard"
grep -rlE "$KEYWORDS" \
~/.cursor/plans/ \
~/.gemini/antigravity/brain/ \
2>/dev/null
3. Assess relevance before copying
Open the artefact. Ask:
- Is the work complete (Cursor: all todos
status: completed; Antigravity: final draft)?
- Does it document a non-trivial architectural decision or reusable pattern?
- Could it inform future work — not just describe what the agent did?
- Does it match the current project context (framework, stack, production state)?
Low-value ephemera (single-file bug fixes, debug logs, one-shot refactors) usually doesn't need saving — the git history already captures the change.
4. Import into the project taxonomy
Ensure the taxonomy exists (first-time setup):
mkdir -p docs/artefacts/by-type/{plans,analyses,validations,research,reports}
mkdir -p docs/artefacts/by-topic/{architecture,deployment,security,frontend,backend}
mkdir -p docs/artefacts/by-agent
Copy the artefact and symlink into topic cross-refs:
cp ~/.cursor/plans/idea-057_text_file_preview_52778d48.plan.md \
docs/artefacts/by-type/plans/IDEA-057-text-file-preview.plan.md
ln -sf ../../by-type/plans/IDEA-057-text-file-preview.plan.md \
docs/artefacts/by-topic/frontend/IDEA-057-text-file-preview.plan.md
cp ~/.gemini/antigravity/brain/<conversation-id>/artifacts/devlog_analysis.md \
docs/artefacts/by-type/analyses/devlog_analysis.md
ln -sf ../../by-type/analyses/devlog_analysis.md \
docs/artefacts/by-topic/security/devlog_analysis.md
5. Unsaved-artefacts audit (run periodically)
#!/bin/bash
PROJECT_KEYWORDS="${1:-myproject|IDEA-|ai_service|dashboard}"
echo "=== 🔍 IDE artefacts potentially related to this project ==="
check_source() {
local label="$1"
local pattern="$2"
for f in $pattern; do
[ -f "$f" ] || continue
if grep -qiE "$PROJECT_KEYWORDS" "$f"; then
basename=$(basename "$f")
if ! find docs/artefacts/ -name "*${basename%.*}*" 2>/dev/null | grep -q .; then
echo " [$label] NOT SAVED: $basename"
fi
fi
done
}
check_source "Cursor" "$HOME/.cursor/plans/*.plan.md"
check_source "Antigravity" "$HOME/.gemini/antigravity/brain/*/artifacts/*.md"
6. Retrieve from the project tree (for downstream work)
Once imported, artefacts are discoverable via simple path queries:
find docs/artefacts/by-agent/researcher/ -name "*.md"
find docs/artefacts/by-type/validations/ -name "*DJANGO*"
find docs/artefacts/by-topic/django-architecture/ -name "*.md"
grep -rli "multi-tenant" docs/artefacts/
Taxonomy Reference
docs/artefacts/
├── by-agent/ # who generated it
│ ├── researcher/
│ ├── test-engineer/
│ └── architect/
├── by-type/ # what kind of artefact
│ ├── plans/ # completed IDE plans (Cursor Plan Mode etc.)
│ ├── analyses/ # exploratory analysis / walkthroughs
│ ├── validations/ # test / validation reports
│ ├── research/ # background research notes
│ └── reports/ # status / audit reports
└── by-topic/ # cross-cutting concern (symlinks into by-type/)
├── architecture/
├── deployment/
├── security/
├── frontend/
└── backend/
by-type/ is the canonical store. by-topic/ holds symlinks for discovery. by-agent/ is optional; skip it if your project doesn't track per-agent provenance.
Example Use Cases
🎯 Project onboarding
New agent session starts, checks docs/artefacts/by-topic/architecture/ for validated patterns before proposing new designs.
🔄 Cross-IDE continuity
Cursor plan for IDEA-057 was generated yesterday; today a Claude Code session picks it up from docs/artefacts/by-type/plans/ and implements against it.
🧪 Production readiness
Before deployment, sweep docs/artefacts/by-type/validations/ for prior sign-offs on related components.
🧹 Periodic cleanup
Weekly run of the unsaved-artefacts audit (§5) captures anything new from Cursor / Antigravity workspaces.
Why This Is Generic
- Assistant-agnostic: Cursor, Claude Code, Antigravity, and future assistants all drop generated work into predictable host locations.
- Project-agnostic: any project can adopt the
docs/artefacts/ taxonomy with a single mkdir -p.
- Format-agnostic: Markdown with YAML frontmatter is universally readable; symlinks are universally resolvable.
- Knowledge preservation: artefacts outlive individual agent sessions and IDE workspaces.
- No lock-in: plain files in the repo, no special tooling required to read them.
References