Anthropic/analytics

2026.10.14.1

Added: The manifest now declares license: Apache-2.0 (SPDX). The registry quality rubric scores a declared license separately from the bundled LICENSE.md.

Upgrade note: Metadata only. No code, schema, or behavioral changes.

2026.10.07.1

Added: collect_cost_by_model method and costByModel resource. Reports per-model token cost over a window from /analytics/cost_report (group_by=model): list vs. paid USD per model, discountUsd/discountPct, optional per-model token-type cost split (groupBy: ["token_type"]), and the unattributed (model-less, e.g. code-execution) total. Issues a second ungrouped query for the same window to obtain authoritative totals, so the per-bucket top-100 group cap becomes detectable via a truncated flag rather than silently under-reporting. Surfaces the export dataRefreshedAt watermark and a rangeExceedsWatermark flag so not-yet-final recent months are explicit. Trusts the API's own post-discount amount and pre-discount list_amount (does not re-apply discountRate). Window spans at most 31 days (the API max); call once per month for longer ranges. The resource also carries totalPaidCents (an exact reconciliation anchor, since per-model USD values are rounded independently and may not sum to totalPaidUsd), modelsFilter (echoes any models[] filter — totals are scoped to it when set, not org-wide), and reconciliationChecked (false when the ungrouped query returns an unexpected dimensioned shape, in which case truncated is set conservatively). The watermark comparison uses numeric instants, not string compare, so non-canonical RFC 3339 timestamps are handled correctly. Additive — existing resources are unaffected.

2026.09.18.1

Upgrade note: Normalized npm:zod dependency version to 4.6.5 across the repo. No behavioral changes in this extension.

2026.10.07.1

Added: collect_cost_by_model method and costByModel resource. Reports per-model token cost over a window from /analytics/cost_report (group_by=model): list vs. paid USD per model, discountUsd/discountPct, optional per-model token-type cost split (groupBy: ["token_type"]), and the unattributed (model-less, e.g. code-execution) total. Issues a second ungrouped query for the same window to obtain authoritative totals, so the per-bucket top-100 group cap becomes detectable via a truncated flag rather than silently under-reporting. Surfaces the export dataRefreshedAt watermark and a rangeExceedsWatermark flag so not-yet-final recent months are explicit. Trusts the API's own post-discount amount and pre-discount list_amount (does not re-apply discountRate). Window spans at most 31 days (the API max); call once per month for longer ranges. The resource also carries totalPaidCents (an exact reconciliation anchor, since per-model USD values are rounded independently and may not sum to totalPaidUsd), modelsFilter (echoes any models[] filter — totals are scoped to it when set, not org-wide), and reconciliationChecked (false when the ungrouped query returns an unexpected dimensioned shape, in which case truncated is set conservatively). The watermark comparison uses numeric instants, not string compare, so non-canonical RFC 3339 timestamps are handled correctly. Additive — existing resources are unaffected.

2026.09.18.1

Upgrade note: Normalized npm:zod dependency version to 4.6.5 across the repo. No behavioral changes in this extension.

2026.09.18.1

Upgrade note: Normalized npm:zod dependency version to 4.6.5 across the repo. No behavioral changes in this extension.

2026.09.15.1

Changed: Bump zod 4.4.3 → 4.6.5

2026.08.28.1

Changed: Normalized the extension license to Apache-2.0 and corrected the copyright holder to "Sean Escriva". Extensions that previously shipped an MIT LICENSE.md are now Apache-2.0, consistent with the repository root and every other extension. No code or behavioral changes.

Upgrade note: License text only. No API, schema, or runtime behavior changed.

2026.08.26.2

Fixed: Restored inline npm:zod@4.4.3 import specifiers so the registry quality scorer can resolve dependencies and score the extension. An earlier release used a bare "zod" import-map specifier, which published but scored as unscored.

Changed: Retained explicit compilerOptions.strict in deno.json. No behavioral or schema changes.

2026.08.28.1

Changed: Normalized the extension license to Apache-2.0 and corrected the copyright holder to "Sean Escriva". Extensions that previously shipped an MIT LICENSE.md are now Apache-2.0, consistent with the repository root and every other extension. No code or behavioral changes.

Upgrade note: License text only. No API, schema, or runtime behavior changed.

2026.08.26.2

Fixed: Restored inline npm:zod@4.4.3 import specifiers so the registry quality scorer can resolve dependencies and score the extension. An earlier release used a bare "zod" import-map specifier, which published but scored as unscored.

Changed: Retained explicit compilerOptions.strict in deno.json. No behavioral or schema changes.

2026.08.26.2

Fixed: Restored inline npm:zod@4.4.3 import specifiers so the registry quality scorer can resolve dependencies and score the extension. An earlier release used a bare "zod" import-map specifier, which published but scored as unscored.

Changed: Retained explicit compilerOptions.strict in deno.json. No behavioral or schema changes.

2026.08.25.1

Changed: Updated labels for improved extension discoverability. Added cross-cutting category labels (security, observability, finops, infrastructure, networking, compliance, devops, ai, incident-response) where applicable.

2026.08.24.1

Added: Output metadata attributes for observability.

• durationMs: Method execution duration in milliseconds.
• collectedBy: Extension name that produced the data.
• fetchedAt: ISO 8601 timestamp when data was fetched (added to resources that previously lacked it).

2026.08.23.1

Changed: Documentation only — no code changes. Documented collect_user_usage in the README, which previously had no usage example at all, and added a ## Troubleshooting section explaining the collected: false degrade pattern shared by the adoption, cost, and userUsage collection paths.

2026.08.21.2

Changed: startDate/endDate on collect_analytics and collect_user_usage are now validated as well-formed YYYY-MM-DD dates before any request is made. Previously a malformed date (e.g. "not-a-date" or "2026-13-99") silently passed the range check — Date parsing of garbage input produces NaN, which made the "end must be after start" comparison a no-op — and the bad value was sent straight to the Analytics API, surfacing only as an opaque upstream 400. Malformed dates now fail fast with a message naming the argument and the value that was rejected.

2026.08.21.1

Changed: Added descriptions to the previously undocumented fields across the snapshot, seats, adoption, cost, and userUsage resource schemas (e.g. total_cents, by_cost_type, collected, byProduct, totals). No behavioral change.

2026.08.14.1

Added: discountRate global arg (fraction, default 0) applying an enterprise volume discount to cost figures derived from the Analytics API's list-price amount fields. Applies to cost.total_cents/total_usd/ by_cost_type and per-user costUsd/totalCostUsd; listCostUsd always stays at list price so the discount and list reference remain comparable. Both cost and userUsage resources gain a discountRate field echoing the rate applied.

Added: totals field on the userUsage resource — grand inputTokens/outputTokens/totalTokens and per-minute rates across all returned users, in the shape @webframp/ai-usage's generic provider registry expects, so this resource can be registered there without a bespoke adapter.

Added: days argument on collect_user_usage (capped at 31, the Analytics API's max window) as an alternative to startDate, so callers (including a future @webframp/ai-usage-scan step) can drive the lookback the same way as the other provider scan methods.

Added: Both collect_analytics and collect_user_usage now reject an inverted date range (endDate at or before startDate) via a shared assertRange check, instead of silently sending a backwards window to the Analytics API — collect_user_usage additionally used this to guard against a near-zero periodMinutes producing wildly inflated per-minute rates in totals.

2026.07.18.1

Added: An upgrades array entry (no-op) to analytics.ts for proper typeVersion tracking on existing instances. No schema or behavior changes.

2026.07.14.1

Added: collect_user_usage — per-user token usage and cost, so you can see who is spending what (and how much of it is Claude Code) rather than only the org aggregate.

• Reads the Enterprise Analytics /v1/organizations/analytics/user_usage_report (tokens: total / output / uncached-input / cache-read, plus requests) and /user_cost_report (amount and list_amount, USD minor units → USD), grouped by product so claude_code is broken out from chat/cowork/etc.

• Optional email filter keeps just one user (matched on actor.email); products filters the surfaces; startDate/endDate bound the window (defaults to the last 30 days; the API caps a window at 31 days).

• Writes a userUsage resource (instance = the sanitized email, or all) with a per-user byProduct breakdown plus totalTokens/totalCostUsd. Reuses the same x-api-key (read:analytics) auth and cursor pagination as collect_analytics. The two reports are collected independently: if only user_cost_report fails (common on seat-based plans, where it 403s while the usage report still serves tokens), the token data is retained with collected: true and costUsd left null — collected: false only when both fail. A single email-filtered result is keyed by the user's unique id (user-<userId>) so emails that sanitize alike cannot collide.

  Caveats: with no email filter the resource stores email + name for every org user (admin-only PII — mind shared datastores); and very large orgs can page-truncate (the filter is applied after fetch), so prefer a date-bounded, filtered query for a single user.

Additive: collect_analytics and its snapshot/seats/adoption/cost resources are unchanged.

2026.07.07.1

Fixed:

• collect_analytics called a non-existent endpoint (/v1/organizations/analytics, HTTP 404) with the wrong param names (start_date/end_date) and a flat-metrics response shape that never matched the real API. Rewritten against the documented Enterprise Analytics API: /analytics/summaries (seats + DAU/WAU/MAU + pending invites, starting_date/ending_date), /analytics/users (paginated, aggregated into adopter counts), and /analytics/cost_report (paginated, summed by cost type).

Added:

• cost resource — token cost/usage over the window (usage-based Enterprise plans).
• Adoption and cost collection are best-effort: a failure in /users or /cost_report no longer fails the whole run (seat-based plans have no cost data).

Changed:

• adoption now reports adopter counts (users with ≥1 project/skill/connector), aggregated from /analytics/users — there is no org-level adoption endpoint.
• The snapshot resource writes to the recent instance (latest is reserved by swamp). Query via data.latest("claude-analytics", "recent").

2026.07.02.1

Added: Initial release. Enterprise Analytics observation model. Collects raw metrics snapshot and extracts structured seat counts (total, active, pending, DAU/WAU/MAU) and adoption metrics (projects, skills, connectors) into separate versioned r

2026.08.14.1

Added: discountRate global arg (fraction, default 0) applying an enterprise volume discount to cost figures derived from the Analytics API's list-price amount fields. Applies to cost.total_cents/total_usd/ by_cost_type and per-user costUsd/totalCostUsd; listCostUsd always stays at list price so the discount and list reference remain comparable. Both cost and userUsage resources gain a discountRate field echoing the rate applied.

Added: totals field on the userUsage resource — grand inputTokens/outputTokens/totalTokens and per-minute rates across all returned users, in the shape @webframp/ai-usage's generic provider registry expects, so this resource can be registered there without a bespoke adapter.

Added: days argument on collect_user_usage (capped at 31, the Analytics API's max window) as an alternative to startDate, so callers (including a future @webframp/ai-usage-scan step) can drive the lookback the same way as the other provider scan methods.

Added: Both collect_analytics and collect_user_usage now reject an inverted date range (endDate at or before startDate) via a shared assertRange check, instead of silently sending a backwards window to the Analytics API — collect_user_usage additionally used this to guard against a near-zero periodMinutes producing wildly inflated per-minute rates in totals.

2026.07.18.1

Added: An upgrades array entry (no-op) to analytics.ts for proper typeVersion tracking on existing instances. No schema or behavior changes.

2026.07.14.1

Added: collect_user_usage — per-user token usage and cost, so you can see who is spending what (and how much of it is Claude Code) rather than only the org aggregate.

• Reads the Enterprise Analytics /v1/organizations/analytics/user_usage_report (tokens: total / output / uncached-input / cache-read, plus requests) and /user_cost_report (amount and list_amount, USD minor units → USD), grouped by product so claude_code is broken out from chat/cowork/etc.

• Optional email filter keeps just one user (matched on actor.email); products filters the surfaces; startDate/endDate bound the window (defaults to the last 30 days; the API caps a window at 31 days).

• Writes a userUsage resource (instance = the sanitized email, or all) with a per-user byProduct breakdown plus totalTokens/totalCostUsd. Reuses the same x-api-key (read:analytics) auth and cursor pagination as collect_analytics. The two reports are collected independently: if only user_cost_report fails (common on seat-based plans, where it 403s while the usage report still serves tokens), the token data is retained with collected: true and costUsd left null — collected: false only when both fail. A single email-filtered result is keyed by the user's unique id (user-<userId>) so emails that sanitize alike cannot collide.

  Caveats: with no email filter the resource stores email + name for every org user (admin-only PII — mind shared datastores); and very large orgs can page-truncate (the filter is applied after fetch), so prefer a date-bounded, filtered query for a single user.

Additive: collect_analytics and its snapshot/seats/adoption/cost resources are unchanged.

2026.07.07.1

Fixed:

• collect_analytics called a non-existent endpoint (/v1/organizations/analytics, HTTP 404) with the wrong param names (start_date/end_date) and a flat-metrics response shape that never matched the real API. Rewritten against the documented Enterprise Analytics API: /analytics/summaries (seats + DAU/WAU/MAU + pending invites, starting_date/ending_date), /analytics/users (paginated, aggregated into adopter counts), and /analytics/cost_report (paginated, summed by cost type).

Added:

• cost resource — token cost/usage over the window (usage-based Enterprise plans).
• Adoption and cost collection are best-effort: a failure in /users or /cost_report no longer fails the whole run (seat-based plans have no cost data).

Changed:

• adoption now reports adopter counts (users with ≥1 project/skill/connector), aggregated from /analytics/users — there is no org-level adoption endpoint.
• The snapshot resource writes to the recent instance (latest is reserved by swamp). Query via data.latest("claude-analytics", "recent").

2026.07.02.1

Added: Initial release. Enterprise Analytics observation model. Collects raw metrics snapshot and extracts structured seat counts (total, active, pending, DAU/WAU/MAU) and adoption metrics (projects, skills, connectors) into separate versioned resources.

2026.07.18.1

Added: An upgrades array entry (no-op) to analytics.ts for proper typeVersion tracking on existing instances. No schema or behavior changes.

2026.07.14.1

Added: collect_user_usage — per-user token usage and cost, so you can see who is spending what (and how much of it is Claude Code) rather than only the org aggregate.

• Reads the Enterprise Analytics /v1/organizations/analytics/user_usage_report (tokens: total / output / uncached-input / cache-read, plus requests) and /user_cost_report (amount and list_amount, USD minor units → USD), grouped by product so claude_code is broken out from chat/cowork/etc.

• Optional email filter keeps just one user (matched on actor.email); products filters the surfaces; startDate/endDate bound the window (defaults to the last 30 days; the API caps a window at 31 days).

• Writes a userUsage resource (instance = the sanitized email, or all) with a per-user byProduct breakdown plus totalTokens/totalCostUsd. Reuses the same x-api-key (read:analytics) auth and cursor pagination as collect_analytics. The two reports are collected independently: if only user_cost_report fails (common on seat-based plans, where it 403s while the usage report still serves tokens), the token data is retained with collected: true and costUsd left null — collected: false only when both fail. A single email-filtered result is keyed by the user's unique id (user-<userId>) so emails that sanitize alike cannot collide.

  Caveats: with no email filter the resource stores email + name for every org user (admin-only PII — mind shared datastores); and very large orgs can page-truncate (the filter is applied after fetch), so prefer a date-bounded, filtered query for a single user.

Additive: collect_analytics and its snapshot/seats/adoption/cost resources are unchanged.

2026.07.07.1

Fixed:

• collect_analytics called a non-existent endpoint (/v1/organizations/analytics, HTTP 404) with the wrong param names (start_date/end_date) and a flat-metrics response shape that never matched the real API. Rewritten against the documented Enterprise Analytics API: /analytics/summaries (seats + DAU/WAU/MAU + pending invites, starting_date/ending_date), /analytics/users (paginated, aggregated into adopter counts), and /analytics/cost_report (paginated, summed by cost type).

Added:

• cost resource — token cost/usage over the window (usage-based Enterprise plans).
• Adoption and cost collection are best-effort: a failure in /users or /cost_report no longer fails the whole run (seat-based plans have no cost data).

Changed:

• adoption now reports adopter counts (users with ≥1 project/skill/connector), aggregated from /analytics/users — there is no org-level adoption endpoint.
• The snapshot resource writes to the recent instance (latest is reserved by swamp). Query via data.latest("claude-analytics", "recent").

2026.07.02.1

Added: Initial release. Enterprise Analytics observation model. Collects raw metrics snapshot and extracts structured seat counts (total, active, pending, DAU/WAU/MAU) and adoption metrics (projects, skills, connectors) into separate versioned resources.

2026.07.07.1

Fixed:

• collect_analytics called a non-existent endpoint (/v1/organizations/analytics, HTTP 404) with the wrong param names (start_date/end_date) and a flat-metrics response shape that never matched the real API. Rewritten against the documented Enterprise Analytics API: /analytics/summaries (seats + DAU/WAU/MAU + pending invites, starting_date/ending_date), /analytics/users (paginated, aggregated into adopter counts), and /analytics/cost_report (paginated, summed by cost type).

Added:

• cost resource — token cost/usage over the window (usage-based Enterprise plans).
• Adoption and cost collection are best-effort: a failure in /users or /cost_report no longer fails the whole run (seat-based plans have no cost data).

Changed:

• adoption now reports adopter counts (users with ≥1 project/skill/connector), aggregated from /analytics/users — there is no org-level adoption endpoint.
• The snapshot resource writes to the recent instance (latest is reserved by swamp). Query via data.latest("claude-analytics", "recent").

2026.07.02.1

Added: Initial release. Enterprise Analytics observation model. Collects raw metrics snapshot and extracts structured seat counts (total, active, pending, DAU/WAU/MAU) and adoption metrics (projects, skills, connectors) into separate versioned resources.

2026.07.02.1

Added: Initial release. Enterprise Analytics observation model. Collects raw metrics snapshot and extracts structured seat counts (total, active, pending, DAU/WAU/MAU) and adoption metrics (projects, skills, connectors) into separate versioned resources.