| name | endor-project-retrieval-bundle |
| description | Use when building a deterministic single-project retrieval bundle via
endor-agent-context: context_manifest.json, dependency explorer artifacts,
optional session summaries, PV index + hydration, optional call-graph export. Not
for namespace topology or breaking-change consumer discovery—hand off to
endor-namespace-relationship-map.
|
| endorlabs | {"catalog":{"workflow_id":"agent-context","cli":"endor-agent-context","module":"endorlabs.workflows.agent_context.cli","default_output":".endorlabs-context/workspace/projects/<slug>_<timestamp>/","agent_visible":true,"composition":"library_api","library_entrypoints":["endorlabs.workflows.agent_context.build_context_manifest","endorlabs.workflows.agent_context.list_package_versions_for_index","endorlabs.workflows.callgraph.run_callgraph_export"]}} |
Project retrieval bundle
Read context_manifest.json first; composition rules in workflow-composition.
Purpose
Produce a versioned, machine-readable retrieval bundle for one project via endor-agent-context (not endor-estate). This is the shared retrieval layer for SBOM slices, DependencyMetadata, optional findings/policies/versions summaries, and call-graph exports — not namespace topology or “who must I warn about a breaking change?”
Vocabulary
| Term | Meaning |
|---|
| Client tenant | --tenant on export — auth context for endorlabs.Client. |
--namespace (optional) | Disambiguates project lookup only; export remains one project. |
--project | Required — single repo (UUID or name). |
| Not namespace scope | Multi-repo / blast-radius → endor-namespace-relationship-map; bulk namespace pull → docs/estate/README.md (user must request). |
Decision table
| Question | Use this skill | Hand off to |
|---|
| SBOM / DM / PV index-hydration for one repo | uv run endor-agent-context | — |
| Findings/policies/repo-version summaries in the bundle | add --session-summaries | endor-retrieve-scan-results for scan RCA |
| Who consumes packages my repo produces (breaking-change blast radius) | — | endor-namespace-relationship-map with --focus-producer-project-uuid, or endor-estate analyze -n <namespace_scope> --only-relationships --focus-producer-project-uuid |
| Full namespace project graph (all edges) | — | namespace relationship map or endor-estate analyze --only-relationships |
| Bulk pull/analyze under a namespace scope (explicit user request) | — | docs/estate/README.md (endor-estate) |
| Function-level call graph search | --callgraph-export + --decode-zstd | endor-fetch-and-search-call-graph — read manifest first; do not decode arbitrary PVs before export |
| Stitched vulnerable-function reachability | bundle as input | endor-reachability-context → endor-reachability-provenance |
Not endor-estate: this skill is always single-project (--project). endor-estate uses namespace scope for multi-project workspace work.
Ordering
- Credentials —
uv run --env-file .env. Never paste secrets into skills or logs.
- Export —
uv run endor-agent-context with --tenant, --project, optional --namespace, --output-dir. Done when context_manifest.json exists. Ambiguous repo URL → --namespace or 24-hex project UUID (AGENTS.md).
- Optional session layer —
--session-summaries → artifacts.session_summaries in the manifest (counts + paths; read summaries only when needed).
- Read manifest — multipass-llm-contract.md for bounds (
inventory.truncated, --pv-limit, session block).
- Downstream — relationship map for cross-repo impact, scan skills for posture, call-graph search as needed.
Multi-pass behavior
| Pass | Role | Default |
|---|
| 1 — Index | package_versions_index.json | On; --no-pv-index to skip |
| 2 — Hydrate | BOM + CG + DependencyMetadata | On unless --index-only |
| 3 — Export | Full PV call-graph export | --callgraph-export (+ --decode-zstd for search) |
Pass 3 writes artifacts.callgraph_export in context_manifest.json. Call-graph export does not populate vector stores — semantic function summaries use tenant VectorStore / VectorStoreQuery / endor-vector-query separately.
Hand off: After Pass 3, follow endor-fetch-and-search-call-graph for path search (direct edges and multi-hop BFS when wrappers sit between source and target).
Bounds: Pass 2 default --pv-limit 5 — do not infer repo-wide dependency/call-graph posture from the summary alone; escalate via manifest warnings or --hydrate-top-n / --hydrate-pv-uuids (multipass-llm-contract.md).
CLI entrypoints
| Step | Path |
|---|
| Export + manifest | endor-agent-context / endorlabs.workflows.agent_context.export |
| Session summaries | --session-summaries |
| PV index helpers | endorlabs.workflows.agent_context.package_versions |
| Call-graph export | --callgraph-export |
Key flags
- Core:
--tenant, --project, --namespace, --output-dir
- Pass 1:
--no-pv-index, --pv-index-max-pages, --pv-index-page-size
- Pass 2:
--index-only, --pv-limit, --dep-metadata-max-pages, --hydrate-pv-uuids, --hydrate-top-n
- Pass 3:
--callgraph-export, --callgraph-max-pages, --decode-zstd
- Session:
--session-summaries
--deterministic — stable ordering for replay/diff
Outputs
- Bundle:
<output_dir>/<slug>_<timestamp>/ with context_manifest.json (version 2), optional index, dependency explorer files, optional callgraph_export/, optional session subdir.
- Stdout: absolute path to
context_manifest.json.
Linked skills
