feat(parsers): add agent-trace.dev v1 sidecar parser#12
Closed
joahg wants to merge 2 commits into
Closed
Conversation
Adds support for the open, vendor-neutral agent-trace.dev v1 spec (https://agent-trace.dev/) as a third parser alongside the existing Claude Code and Codex/Copilot native log parsers. Why --- ai-blame previously had to grow one bespoke parser per agent. The agent-trace.dev v1 spec is an open, agent-agnostic JSON sidecar format designed exactly for this attribution use case, and any producer of the spec (first-party emitter or third-party exporter from native logs) is unlocked at once. What ---- - New `src/parsers/agent_trace.rs` reads `.agent-trace/*.json` records conforming to the v1 schema and emits one `EditRecord` per `files[].conversations[].ranges[]` entry. - Registered in `ParserRegistry::new()` first so the unambiguous `.json` + spec-shaped discriminator wins before the `.jsonl` parsers see the file. - `get_agent_trace_dirs()` discovers `<cwd>/.agent-trace/` and `~/.agent-trace/` automatically and feeds them into `get_all_trace_dirs()`. - Smoke-tested end-to-end against 21 real spec records: 126 edits across 41 files extracted via `ai-blame stats` / `report`. Limitations ----------- agent-trace.dev records carry attribution ranges + `content_hash` (spec §6.3) rather than raw `old_string` / `new_string` patches, so `stats` / `timeline` / `report` / `transcript` work fully but `blame` / `annotate` cannot perform the same patch-walk reconstruction as the native parsers. Documented in the parser module docstring and in the README. Tests ----- 7 new unit tests in `parsers::agent_trace::tests` covering parse, filter, fallback, can_parse acceptance/rejection, and `collect_trace_files` extension filtering. Full `cargo test` is green (22 passing); `cargo fmt --check` and `cargo clippy` clean. Amp-Thread-ID: https://ampcode.com/threads/T-019e0ee7-99b6-72e7-a9eb-e71bba013ceb Co-authored-by: Amp <amp@ampcode.com>
joahg
force-pushed
the
joah/agent-trace-dev-parser
branch
from
67c80d4 to
63a36fd
Compare
When the spec's top-level `tool` block is absent (some emitters drop
the whole block because the spec requires both `name` and `version`
when `tool` is present, and they only have a name) we were falling all
the way back to `agent_tool=agent-trace` / `model=unknown`, even
though the conversation URL or related-session URN clearly identified
the agent.
Now we walk a per-conversation derivation chain:
- `agent_tool` ← `tool.name` → conversation URL host (ampcode.com →
amp, cursor.{sh,com} → cursor, claude.ai → claude-code,
{openai,chatgpt}.com → codex, *.block.xyz → goose) → session URN
agent slug (`urn:*:session:<agent>:<id>`) → `agent-trace`.
- `model` ← per-range `contributor.model_id` → conversation
`contributor.model_id` → `<contributor.type> (model unspecified)`
(e.g. `ai (model unspecified)`) → `unknown`.
- `session_id` ← trailing path/URN segment of conversation `url` →
trailing segment of `related[type=session]` → full URL/URN →
`unknown`. Trailing-segment matches what the native parsers use, so
cross-tool correlation works.
Also derives per-conversation rather than per-record so a single
record covering multiple agents (e.g. a hand-off) attributes each
conversation correctly.
Verified end-to-end against the same 21 real records: previously every
edit displayed as `agent-trace / unknown`; now they correctly
attribute as `amp / ai (model unspecified)`.
Tests: 4 new (agent_url_host_mapping, agent_session_urn_extraction,
derives_agent_from_amp_url_when_tool_block_absent,
record_tool_name_takes_priority_over_url_sniffing); existing tests
updated for trailing-segment session_id.
cargo test: 26 passing; fmt + clippy clean.
Amp-Thread-ID: https://ampcode.com/threads/T-019e0ee7-99b6-72e7-a9eb-e71bba013ceb
Co-authored-by: Amp <amp@ampcode.com>
Author
|
🤖 Sent by Joah's AI agent: End-to-end output against 21 real
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Adds a third parser,
AgentTraceParser, that consumes the openagent-trace.dev v1 spec (
.agent-trace/*.jsonsidecars) and turns each
files[].conversations[].ranges[]entry intoan
EditRecord. Registered inParserRegistry::new()ahead of theexisting Claude / Codex parsers and auto-discovered via a new
get_agent_trace_dirs()(<cwd>/.agent-trace/,~/.agent-trace/).Why
ai-blame today supports Claude Code and Codex/Copilot via bespoke
parsers per agent. The agent-trace.dev v1 spec is an open,
agent-agnostic JSON sidecar format intended exactly for this
attribution use case, so a single parser unlocks support for every
producer of the spec — first-party emitters or third-party exporters
that convert an agent's native logs — instead of growing the
parser-per-agent matrix forever.
End-to-end smoke test
Ran against 21 real spec records validated against the upstream
v1 schema:
Limitations (documented in the parser docstring + README)
agent-trace.dev records carry attribution ranges + content_hash
(spec §6.3) rather than raw
old_string/new_stringpatches. As aresult:
stats,timeline,report,transcriptblame,annotateA future iteration could use
content_hashfor position-independentattribution — left as follow-up.
Mapping
Each spec range becomes one
EditRecord:file_path—files[].path(already repo-relative per spec).timestamp— record's top-leveltimestamp(per-revision, not per-edit).model—contributor.model_id(range override → conversation default →unknown).session_id—conversation.url→related[type=session].url→unknown.agent_tool/agent_version— record's top-leveltoolblock.is_create/change_size— heuristic: ranges starting at line 1 are treated as create-shaped;change_sizeis the line span.Tests
7 new unit tests in
parsers::agent_trace::testscovering parseoutput, file-pattern filter, session-id fallback to
related[],can_parseacceptance + rejection (incl. refusing to steal.jsonlfiles from sibling parsers), and
collect_trace_filesextensionfiltering.
cargo fmt --checkandcargo clippy --libare both clean.