Repository navigation
Cloudseed v0.2.3
Cloudseed now records usage for its own agent tasks and MCP calls, with the same report available through the CLI, agents, MCP and web console. Agent report review also works when cs or cloudseed was not originally on PATH.
What changed
- Cloudseed-only usage: run
cs usage, call MCPcloudseed_usage, or open Agents & MCP → Cloudseed usage. Reports show task status, available token counts, model provenance and explicit reasons for missing measurements. Filters, pagination and JSON export are available. - Optional ccusage analysis:
cs usage installinstalls the pinned, verified ccusage 20.0.24 native analyzer.cs usage report --engine ccusageestimates cost from Cloudseed-owned metadata and offline prices. It does not import unrelated conversations or search global agent history. - Simpler agent execution: the separate compression proxy, its runtime dependencies and its controls have been removed. Cloudseed uses the agent's normal transport. The optional Context brief is Cloudseed's task context summary.
- Reliable command access: private
csandcloudseedlaunchers point to the exact running installation, with an absolute path fallback. - Complete saved-report review:
cs evidence list/readand MCPcloudseed_evidenceprovide redacted reports and diagnostic logs through revision-bound pages. Agents can read every finding, observation timestamp, coverage limit and next action without starting another scan. - More useful scan diagnostics: new cloud scans retain bounded, redacted diagnostic excerpts with omission counts. Zero-error counters and ordinary prose mentioning errors no longer create false incomplete verdicts.
- More resilient Mac signing: bounded retries handle recognized transient Apple timestamp-service failures; publisher, hardened runtime and secure timestamp checks remain required.
Start using it
After installing v0.2.3, restart the Cloudseed console and MCP server so they use the new executable. New Cloudseed-launched agent tasks populate the usage report:
cs usage
cs usage install
cs usage report --engine ccusageThe native report works without installing the optional analyzer. Existing sessions cannot retroactively supply missing measurements. MCP reports calls, duration and response sizes; model tokens remain unavailable because the MCP client does not send them to Cloudseed. Cost estimates are neither invoices nor subscription-quota measurements.
Read the usage guide, agentic guide, MCP report workflow, and updated scenario 14.
Validation
- Release workflow: all 10 jobs passed at commit
14adc694c2d2c75a2e5dbb12f53fbcda707b4fe9. - Source suite: 3,960 tests in each of six Linux/macOS and Python combinations, with explicit optional/platform skips. All 19 offline/local scenarios passed (725 checks; 62 live/credentialed skips). Terraform: seven roots validated, 124 mocked cases passed.
- Every released binary passed 25 runtime tests plus five real native ccusage tests; both containers passed 23 plus five. Release-platform scenario checks totaled 1,500, with 60 explicit skipped-action occurrences.
- Independently downloaded all 22 assets; verified checksums, manifests, inventories, all 16 file attestations and both container attestations against the exact source/workflow/tag.
- Both downloaded Mac binaries passed strict Developer ID signature checks for Nimesh Pandeya, team
QPF2VF2885, hardened runtime and secure timestamps. - The downloaded Apple Silicon binary passed seven targeted usage/report tests and a real Claude client exercise using a synthetic local API. Complete report paging, redaction and exact recorded token counts passed. Its own release verifier also returned PASS.
- Updated GitHub Pages deployed from the same source; seven live endpoints and 54 content checks passed.
Automated source, package and offline scenario checks do not establish live deployment or recovery acceptance across all clouds. Developer ID signing is separate from Apple notarization; this release does not claim notarization.