ccusage Agent Sources

Every agent is an adapter crate in rust/adapters/<agent> that turns one source's raw logs into the shared report shapes. ccusage <report> aggregates all of them; ccusage <agent> <report> scopes to one.

Which reports an agent accepts is not uniform, and it is encoded twice: STANDARD_AGENT_REPORTS / OPENCODE_AGENT_REPORTS in rust/crates/ccusage-cli/src/types.rs drive parsing, while agent_report_supported in rust/crates/ccusage-cli-parser/src/parser.rs decides what is accepted per agent name. Changing an agent's report set means updating both. Keep command names and flag semantics aligned across agents unless the source data forces the difference, and say so in a comment when it does.

Cost handling is shared, not per-agent: auto, calculate, and display differ only in how a record's precomputed costUSD competes with token-derived cost, decided in rust/crates/ccusage-core/src/cost.rs. Pricing comes from LiteLLM with an embedded snapshot behind --offline (pricing.rs, and the rust skill for how both snapshots are built into the binary).

Before Changing A Source

Read that agent's own docs first — they carry the paths, record shapes, token mapping, model fallbacks, and cost quirks that are not shared:

• rust/adapters/<agent>/README.md - what the crate owns and where its data lives.
• rust/adapters/<agent>/src/README.md - record shapes and token/cost rules, present for the sources that need them.

Architecture and the shared-vs-source boundary live in rust/adapters/README.md; the per-change workflow in rust/adapters/AGENTS.md.