feat(source): add Augment Code adapter

[ayagmar]

Objective

Add an augment source adapter for local Augment Code session files so Augment usage can flow through the existing reporting pipeline without introducing any auth or editor-integration logic.

Why this looks implementable

The local data under ~/.augment/ appears to contain a parseable session format already:

• session files live under ~/.augment/sessions/*.json
• each session has top-level metadata such as sessionId, created, modified, and agentState.modelId
• chatHistory[] contains exchanges with request_nodes and response_nodes
• per-turn usage signal appears on response_nodes entries with type: 10 and a populated token_usage
• repo / cwd attribution appears in request_nodes entries with ide_state_node.current_terminal.current_working_directory and workspace_folders[].repository_root
• provider can appear in sibling response metadata (for example metadata.provider on nearby response nodes)

That is enough to support a first local adapter behind the current SourceAdapter boundary.

Source shape to support

• directory-backed source under ~/.augment/sessions/*.json
• one JSON file per session
• each exchange can contain multiple response nodes, but usage should be emitted only from the token-usage node(s)
• model is available at session level via agentState.modelId
• explicit cost does not appear to be present in the session files, so cost should stay estimated unless a better local source is identified later

Implementation plan

1. Adapter file

   • add src/sources/augment/augment-source-adapter.ts
   • implement SourceAdapter with id = 'augment'

2. Discovery

   • discover only ~/.augment/sessions/*.json
   • keep this source directory-backed and selectable through generic --source-dir augment=...
   • validate explicit directory overrides consistently with the existing directory-backed adapters

3. Parsing

   • parse one session JSON file at a time
   • iterate chatHistory[] exchanges and emit usage events from the response node that carries token_usage
   • use the token-usage node timestamp (timestamp_ms) as the event timestamp, with fallback to top-level session timestamps when required
   • resolve model from agentState.modelId
   • resolve provider from sibling response metadata when present; otherwise leave it unset or map conservatively from the model only if the data is clearly reliable
   • resolve repo attribution from request-side IDE state, preferring:
     • ide_state_node.current_terminal.current_working_directory
     • then workspace_folders[].repository_root
   • map local usage buckets carefully into createUsageEvent
     • input_tokens
     • output_tokens
     • cache_read_input_tokens
     • cache_creation_input_tokens
   • do not invent explicit cost from auth/session files
   • ignore non-usage response nodes (thinking, final text nodes, tool-use nodes) and malformed exchanges without aborting the whole file
   • use parseFileWithDiagnostics() so malformed files / invalid timestamps can surface skipped-row reasons

4. Token semantics follow-up to settle during implementation

   • the local token_usage object also includes fields such as:
     • system_prompt_tokens
     • chat_history_tokens
     • current_message_tokens
     • tool_definitions_tokens
     • tool_result_tokens
     • assistant_response_tokens
   • decide and document whether these should remain informational-only or contribute to canonical billable buckets / total tokens
   • keep the first implementation simple and deterministic; if some counters are overlapping views rather than additive buckets, avoid double counting

5. Wiring

   • register augment in src/sources/create-default-adapters.ts
   • update supported-source IDs and stable-order tests/help output
   • keep the source on generic --source-dir support; no bespoke flag needed

6. Tests and docs

   • add tests/sources/augment-source-adapter.test.ts
   • update shared registry/help tests in:
     • tests/sources/create-default-adapters.test.ts
     • tests/cli/create-cli.test.ts
   • update README.md if the source is added to the supported-source list

Code paths to touch

• src/sources/augment/augment-source-adapter.ts
• src/sources/create-default-adapters.ts
• src/sources/parsing-utils.ts
• src/sources/parse-diagnostics.ts
• src/utils/discover-files.ts
• src/domain/usage-event.ts
• tests/sources/augment-source-adapter.test.ts
• tests/sources/create-default-adapters.test.ts
• tests/cli/create-cli.test.ts
• README.md

Verification

• parser fixtures cover:
  • session-level model extraction
  • per-exchange token-usage extraction
  • repo-root / cwd extraction from IDE-state request nodes
  • provider extraction from sibling response metadata when present
  • invalid/missing timestamps and malformed exchanges surfaced via diagnostics
  • non-usage nodes ignored without creating duplicate events
  • stable handling of the extra prompt/context counters without double counting
• llm-usage daily --source augment --source-dir augment=<fixture-dir> works
• supported-source help output and registry tests are updated
• full validation passes:
  • pnpm run lint
  • pnpm run typecheck
  • pnpm run test
  • pnpm run format:check