feat(kernelforge): export-runs reads AgentKernelArena driver.log sessions by amd-ama10002-2 · Pull Request #1950 · AMD-AGI/Hyperloom

amd-ama10002-2 · 2026-10-16T20:53:49Z

Motivation

We are building knowledge-base support for KernelForge runs (kernel writing): turning KernelForge's raw logs into structured Experiences in Hyperloom's Experience KB. We are also changing the logs KernelForge writes so they carry what the KB extracts: planner plans open with a Summary: line, and implementers keep a <plan>.reasoning.md work log that opens the same way. This PR teaches kernelforge export-runs the AgentKernelArena driver.log session.

Stack: merge in this order, one at a time. Each PR targets the branch of the PR above it in this list.

feat(kernelforge): plans open with a Summary line; implementers keep <plan>.reasoning.md #1949 — Summary line in plans; implementer reasoning logs
feat(kernelforge): collect plain forge-loop runs as Experiences with export-runs and kernelforge-v1 #1944 — export-runs and the kernelforge-v1 mapping, for plain forge-loop runs
feat(kernelforge): export-runs reads AgentKernelArena driver.log sessions #1950 — export-runs reads AgentKernelArena driver.log sessions (this PR)
feat(kernelforge): export-runs reads AgentKernelArena campaign logs #1951 — export-runs reads AgentKernelArena campaign logs

Don't collect AgentKernelArena logs into a KB until #1951 lands: until then a campaign's artifacts are still read as a plain forge-loop run.

What changes

kernelforge export-runs reads an AgentKernelArena session directory (driver.log and arena_logs/ beside the task's directory, as the forge template writes for forge-rewrite-by-flydsl): one entry for the PORT step and one per forge-loop iteration, under the launcher's run name. Before, such a session yielded no entry. Collection makes an Experience of each iteration but skips the PORT step's entry, with a reason saying so: forge-rewrite logs no reasoning for the port, and every Experience requires reasoning.
Its identity comes from the task's own files and the session's config.yaml, its provenance from the session header: the KernelForge and arena checkouts, the model, and the KB mode.
rationale.reasoning taken from KernelForge's Rationale line leaves out the in-session gate's [gate edits=...] prefix, and provenance.kb_record is set only when forge-rewrite's KB publication names a written record, never its failure reason.
Refactor-only parts: the gate-prefix rule moves from forge_loop.py into archive.py (rationale_text), so the plain-run and arena readers share it.

Known limitations

Decision labels are read from the line shapes KernelForge's runner prints (as of 1c14e79e1); a change to those messages needs a matching change here.
A session directory must hold exactly one arena log, one arena run directory and one task directory; otherwise it is one entry saying why.
Arena log times are naive local times too, read as UTC.

Tests: new test_run_export_arena.py, session layout.
Size/complexity triggers crossed: session_log._line (cyclomatic 11) and _label (12); each is one flat dispatch over the line shapes KernelForge prints.
Observable effect: see What changes.
Breaking changes: no
PR addresses single concern: yes

github-actions · 2026-10-16T20:55:25Z

Code metrics gate: PASSED

Judged: the 11 files this change touches. Files it does not touch never fail it; anything found there is listed under Outside this change for information.

Dimension	Limit	Baseline entries	vs base
Cyclomatic complexity	fails > 20	68	+0
Cognitive complexity	fails > 30	274	+0
Function length (lines)	fails > 80	412	+0
Nested block depth	fails > 5	25	+0
Parameter count	fails > 5	95	+0
Module length (lines)	warns > 800, fails > 1200	51	+0
Duplicated lines per file	fails > 0	97	+0
Dead code (vulture)	fails > 0	2	+0

Check	Findings	Effect
Comments	0
English only	0
Production code imports test code	0
Repeated literals	0
A1 Private name of another package	0
A8 Flag argument	0
A9 Tuple return	0
A10 Raw vocabulary comparison	0
A11 Hidden global write	0
A12 Pure forwarder	0
A12 Single-implementation abstraction	0

105 findings in files this change does not touch

Outside this change (informational): 105

Unit	Dimension	Baseline	Now
`src/hyperloom/orchestrator/enablement/recipe/projections.py` `project_launch_evidence`	Cognitive complexity	absent	36
`src/hyperloom/orchestrator/state/shared_state.py` `resolve_graded_comparison`	Cognitive complexity	absent	34
`src/hyperloom_kb/collect/mapping.py` `_experience`	Cognitive complexity	absent	31
`src/hyperloom_kb/schema.py` `FieldDeclaration.__post_init__`	Cognitive complexity	absent	32
`src/kernelforge/cli/forge_loop_command.py` `_prepare_measurement_driver`	Cognitive complexity	absent	36
`src/kernelforge/cli/forge_loop_command.py` `_read_kb_warm_start`	Cognitive complexity	absent	40
`src/kernelforge/loop/runner.py` `IterationLoop._collect_session_candidate`	Cognitive complexity	absent	44
`src/kernelforge/loop/runner.py` `IterationLoop._rule_on_attempt`	Cognitive complexity	absent	38
`src/hyperloom/inference_optimizer/cli/__init__.py` `_reexport_session_environment`	Function length (lines)	absent	111
`src/hyperloom/inference_optimizer/cli/__init__.py` `_resolve_agent_lineup`	Function length (lines)	absent	89
`src/hyperloom/inference_optimizer/cli/__init__.py` `_resolve_run_topology`	Function length (lines)	absent	92
`src/hyperloom/inference_optimizer/cli/__init__.py` `_resume_session`	Function length (lines)	absent	170
`src/hyperloom/inference_optimizer/cli/__init__.py` `_seed_fresh_session`	Function length (lines)	absent	170
`src/hyperloom/inference_optimizer/multi_node/cli.py` `_restart_multinode`	Function length (lines)	absent	107
`src/hyperloom/inference_optimizer/multi_node/scripts/launch_multinode.py` `_build_parser`	Function length (lines)	absent	84
`src/hyperloom/inference_optimizer/multi_node/scripts/launch_multinode.py` `_spawn_rank_actors`	Function length (lines)	absent	84
`src/hyperloom_kb/sync.py` `GlobalSync.pull`	Function length (lines)	absent	84
`src/kernelforge/cli/forge_loop_command.py` `_RunReporter.build_result`	Function length (lines)	absent	91
`src/kernelforge/cli/forge_loop_command.py` `_prepare_measurement_driver`	Function length (lines)	absent	136
`src/kernelforge/cli/forge_loop_command.py` `_read_kb_warm_start`	Function length (lines)	absent	149
`src/kernelforge/loop/runner.py` `IterationLoop._close_iteration`	Function length (lines)	absent	117
`src/kernelforge/loop/runner.py` `IterationLoop._collect_session_candidate`	Function length (lines)	absent	184
`src/kernelforge/loop/runner.py` `IterationLoop._implementer_history`	Function length (lines)	absent	112
`src/kernelforge/loop/runner.py` `IterationLoop._rule_on_attempt`	Function length (lines)	absent	102
`src/kernelforge/loop/runner.py` `IterationLoop._source_attempt`	Function length (lines)	absent	169

...and 80 more; the job log lists every row.

Verdict computed by the gate scripts in /home/runner/work/_temp/code-metrics-gate, not by this tree's copy.

_{Thresholds and sources: [tool.hyperloom.code_metrics] in pyproject.toml. Tools: complexipy 8.0.1, jscpd 5.4.1, ruff 0.16.2, vulture 2.16.}

Copilot

🟡 Changes recommended

Unresolved parsing errors can omit decisions and measurements or export incorrect reasoning and KB record references.

4 open issues

What changed in this PR

Extends kernelforge export-runs to read AgentKernelArena driver-log sessions for Experience KB collection. Campaign-log support remains deferred to #1951.

Changes:

Exports PORT steps and forge-loop iterations from session logs.
Reads task identity and session provenance.
Adds session tests and updates export documentation.

File	Description
`src/kernelforge/tests/test_run_export_arena.py`	Tests session layouts, decisions, and failures.
`src/kernelforge/run_export/session_log.py`	Parses session timelines and exports steps.
`src/kernelforge/run_export/forge_loop.py`	Documents the shared timezone limitation.
`src/kernelforge/run_export/document.py`	Integrates arena session discovery.
`src/kernelforge/run_export/command.py`	Updates CLI help for session support.
`src/kernelforge/run_export/arena.py`	Reads session layout, identity, and provenance.
`docs/reference/experience-kb-collect.md`	Documents arena collection inputs.
`docs/kernelforge/reference/cli.md`	Describes expanded export support.

_{Review effort: Balanced}

amd-ama10002-2 · 2026-10-16T23:19:53Z

/review

github-actions · 2026-10-16T23:20:25Z

Automated review started for head 0ea198d. Progress: https://github.com/AMD-AGI/Hyperloom/actions/runs/38980848102

github-actions · 2026-10-16T23:21:55Z

Automated review: no verdict applied -- the review job ended with result 'failure'. Comment /review to retry. Run: https://github.com/AMD-AGI/Hyperloom/actions/runs/38980848102

amd-ama10002-2 · 2026-10-17T01:26:50Z

This is the result of running the agent review-pr skill locally (the /review skill I think)

PR #1950 -- feat(kernelforge): export-runs reads AgentKernelArena driver.log sessions

What it does: kernelforge export-runs had no reader for an AgentKernelArena driver.log session, so a forge-rewrite session's PORT step and forge-loop iterations never reached hyperloom-kb-collect. This PR adds that reader (run_export/arena.py, run_export/session_log.py) beside the plain forge-loop one, and keeps the in-session gate's [gate edits=...] prefix out of reasoning and forge-rewrite's failure reasons out of kb_record.

Blocking issues: 2

[X3] The reference doc promises one Experience per PORT step; collection skips every PORT step [verified]
Problem: docs/reference/experience-kb-collect.md:219-220 now says the kernelforge-v1 mapping gives "one Experience per forge-rewrite PORT step", but port_unit never writes reasoning, transcripts are linked only to plain runs (forge_loop_runs, document.py:33), and the mapping requires reasoning (kernelforge-v1.yaml:53). Running test_run_export_arena.py's own session (SETUP, KEPT, MERGED, UNFINISHED) through normalize() and hyperloom-kb-collect --mapping kernelforge-v1 --dry-run collects iteration 1 and skips sess-gemm/port with "No reasoning, which every Experience requires". The skip sentence at :256-258 names only a merge as the case that never has reasoning, and the one test that collects a PORT unit (test_collect_kernelforge.py:127-150) hand-builds a reasoning field the exporter never writes. src/kernelforge/run_export/session_log.py:251
Impact: no driver.log session ever yields a PORT Experience although the reference doc says it does, and the skip reason's advice to provide logs that record the agent's reasoning cannot be followed, since nothing the export reads gives a PORT step reasoning.
Action: Either give port_unit a reasoning source the export reads, or state in experience-kb-collect.md:219-220 and :256-258, and in the description, that a PORT step is exported but never collected, correcting the same sentence in kernelforge-v1.yaml:3 and kernelforge-run-v1.yaml:1.
[C3] stop_line is written for every undecided iteration and read by nothing [verified]
Problem: _stop adds stop_line to each iteration that has no decision line, and nothing under src/ reads it: it is the only key the session layout writes that kernelforge-v1.yaml never names, collect's skip report carries only unit_id and reason, no doc mentions it, and only test_run_export_arena.py:201,306 assert it. The description does not declare it record-only. src/kernelforge/run_export/session_log.py:413
Impact: runs.json gains a field that never reaches an Experience or the collect report and is kept in step with _stop only for its own tests.
Action: Delete stop_line and its two assertions, or list it in docs/kernelforge/reference/cli.md as a record-only field and say what an operator reads it for.

SKIPPED: Step 7 independent reader -- no second reader was available (two attempts to start one were aborted), so both findings above were refuted by this reviewer only.
SKIPPED: T1 real-artifact check -- no AgentKernelArena session from an actual run is in the tree or was available to this review, and no in-tree doc describes its layout, so the directory shape (arena_ws_*/run_*/Aiter-task_*, .task-sessions/) and the arena log line prefix the fixtures assume are unchecked; KernelForge's own line shapes were checked against the print statements that produce them.
SKIPPED: V4 -- "# PARTIAL: #1032 changes 1075 files, only its first 100 were compared", overlap there unchecked
SKIPPED: V4 -- "# PARTIAL: #1154 changes 1232 files, only its first 100 were compared", overlap there unchecked
SKIPPED: V4 -- "# PARTIAL: #1276 changes 1311 files, only its first 100 were compared", overlap there unchecked
SKIPPED: V4 -- "# PARTIAL: #1675 changes 161 files, only its first 100 were compared", overlap there unchecked
SKIPPED: V4 -- "# PARTIAL: #1869 changes 331 files, only its first 100 were compared", overlap there unchecked
SKIPPED: V4 -- "# PARTIAL: #1922 changes 118 files, only its first 100 were compared", overlap there unchecked

Checked: src/kernelforge/run_export/{arena,session_log,archive,forge_loop,document,command}.py, test_run_export_arena.py, test_run_export.py, hyperloom_kb test_collect_kernelforge.py, mappings/kernelforge-v1.yaml, declarations/kernelforge-run-v1.yaml, docs/reference/experience-kb-collect.md, docs/kernelforge/reference/cli.md, rewrite_by_flydsl/{runner,kb,report,optimize}.py, loop/runner.py, knowledge/kb_store/identity/kernel_recipe.py, .importlinter | Ran: normalize() then hyperloom-kb-collect --mapping kernelforge-v1 --dry-run on the test's session at head; a sweep of every key that session writes against the mapping's $run. names; ruff check --select C901; git grep sweeps at head and base; CI at head re-read before posting (33 success, 2 skipped main-only jobs) | Base: 7a1171a | Head: 3acafc2

amd-ama10002-2 · 2026-10-17T01:41:14Z

This is the result of running the agent review-pr skill locally (the /review skill I think)

PR #1950 -- feat(kernelforge): export-runs reads AgentKernelArena driver.log sessions

What it does: kernelforge export-runs gains a reader for the AgentKernelArena driver.log session (driver.log and arena_logs/ beside the task's directory, as forge-rewrite-by-flydsl writes it), so a session yields one runs.json entry for its PORT step and one per forge-loop iteration for hyperloom-kb-collect --mapping kernelforge-v1; before, it yielded none. The head commit also keeps the in-session gate's [gate edits=...] prefix out of exported reasoning and sets provenance.kb_record only when forge-rewrite's publication names a written record.

Blocking issues: 1

[X3] The reference doc promises an Experience per PORT step, but kernelforge-v1 skips every PORT step this layout exports [verified]
Problem: docs/reference/experience-kb-collect.md:219-220 now says the mapping gives "one Experience per forge-rewrite PORT step", but port_unit (src/kernelforge/run_export/session_log.py:251) writes no reasoning, the session layout links no transcript, and src/hyperloom_kb/mappings/kernelforge-v1.yaml:53 requires $run.reasoning; the skip sentence at experience-kb-collect.md:256-258 names only a merge as the case that never has reasoning. Reproduced at 3acafc2 with test_run_export_arena.py's own PORT OK session: normalize() writes sess-gemm/port, and hyperloom-kb-collect --mapping kernelforge-v1 --dry-run skips it with "No reasoning, which every Experience requires ..." while collecting sess-gemm/iteration-1. The one test that collects a PORT unit (src/hyperloom_kb/tests/test_collect_kernelforge.py:127, from feat(kernelforge): collect plain forge-loop runs as Experiences with export-runs and kernelforge-v1 #1944) hand-adds a reasoning value the exporter never writes.
Impact: no forge-rewrite PORT step reaches the KB, so the declaration's PORT OK / PORT FAILED outcomes never occur, while the reference doc says they do; the skip reason's remedy ("Provide logs that record the agent's reasoning") cannot be followed, since forge-rewrite logs no rationale for its port and the export reads no transcript for a session.
Action: Decide whether a PORT step should be collected. If it should, give the PORT unit a reasoning source the export reads, or exempt kind: port from the reasoning requirement in kernelforge-v1.yaml:53. If it should not, say at experience-kb-collect.md:219-220 and :256-258, and in the description's first What-changes bullet, that a PORT step is exported but never collected, and correct the same sentence in kernelforge-v1.yaml:3 and kernelforge-run-v1.yaml:1 (feat(kernelforge): collect plain forge-loop runs as Experiences with export-runs and kernelforge-v1 #1944).

Checked: src/kernelforge/run_export/{arena,session_log,archive,forge_loop,document,command}.py; docs/kernelforge/reference/cli.md; docs/reference/experience-kb-collect.md; src/hyperloom_kb/mappings/kernelforge-v1.yaml; src/hyperloom_kb/declarations/kernelforge-run-v1.yaml; src/hyperloom_kb/tests/test_collect_kernelforge.py; the producers' prints in src/kernelforge/rewrite_by_flydsl/{runner,optimize,kb,report}.py and src/kernelforge/loop/runner.py; src/kernelforge/knowledge/kb_store/identity/kernel_recipe.py; src/kernelforge/orchestrator/agent.py _gate_outcome_tag; the full file lists of every open PR | Ran: radon cc on the added functions; git grep sweeps of each new and changed name over src/ and scripts/; normalize() plus hyperloom-kb-collect --mapping kernelforge-v1 --dry-run on the arena test's session; the head commit's new tests on 7e4cf6b (3 fail) and 3acafc2 (pass); CI at head (all green) | Base: 7a1171a | Head: 3acafc2

AgentKernelArena's forge template runs KernelForge's forge-rewrite-by-flydsl pipeline in a session directory holding driver.log and arena_logs/, beside the task's directory. export-runs read none of it, so those runs gave the KB nothing. Such a session is now one run: a unit for forge-rewrite's PORT step (seq 0) and one per forge-loop iteration, under the launcher's run name as run_id. session_log reads the arena's timestamped log and driver.log as that timeline: an iteration's fields come from its own lines (header, merge text, the agent's rationale, what it measured, and the decision line KernelForge printed). The task's conditions come from the task's own files, and the session header gives the KernelForge and arena checkouts, the model and the KB mode. The files a session holds are not read again as a plain forge-loop run. Observable effect: `kernelforge export-runs` on a directory holding AgentKernelArena driver.log sessions writes their PORT step and iterations, which kernelforge-v1 collects as Experiences; before, those sessions yielded no unit. Co-authored-by: Cursor <cursoragent@cursor.com>

…ications out of arena Experiences Two things the AgentKernelArena session reader took from KernelForge's log were not what they looked like. A gate-enabled implementer's Rationale line opens with the in-session gate's outcome, "[gate edits=... pass=... end=...]", which became the first words of rationale.reasoning. The plain-run reader already removed it; the rule now lives once in run_export.archive (rationale_text), and both readers use it. forge-rewrite prints "KB publish: <reason or solution>", so a failed write logs its reason where a successful one logs the record: "not_configured" became provenance.kb_record, and a multi-word error its first word. A payload now counts as a record only when it has the KB store's record shape, <canonical id>/<session id>, built from the store's own scheme name and dimension count, so logs already written are read correctly too. The merge fixture now carries the [bench] line a real merge prints before its decision line, and the test asserts its score. Observable effect: an arena Experience's rationale.reasoning no longer starts with the gate's telemetry, and provenance.kb_record is set only for a publication that wrote a record. Co-authored-by: Cursor <cursoragent@cursor.com>

…ich logs no reasoning export-runs writes an entry for a session's PORT step, but forge-rewrite logs no reasoning for the port and kernelforge-v1 requires reasoning, so collection always skips that entry. The collection reference, the mapping's description and the declaration's header said there is one Experience per PORT step; they now say there is one per forge-loop iteration and name the PORT step among the units that never have reasoning, as the mapping's no-reasoning skip reason now does.

This was referenced Oct 16, 2026

feat(kernelforge): export-runs reads AgentKernelArena campaign logs #1951

Open

feat(kernelforge): collect plain forge-loop runs as Experiences with export-runs and kernelforge-v1 #1944

Open

amd-ama10002-2 mentioned this pull request Oct 16, 2026

feat(kernelforge): plans open with a Summary line; implementers keep <plan>.reasoning.md #1949

Open

amd-ama10002-2 added the run-forge-e2e Run the single-GPU Forge e2e smoke on this PR (again on every new commit while set) label Oct 16, 2026

amd-ama10002-2 requested a balanced review from Copilot October 16, 2026 23:04

Copilot started reviewing on behalf of amd-ama10002-2 October 16, 2026 23:06

View session

Copilot AI reviewed Oct 16, 2026

View reviewed changes

Comment thread src/kernelforge/run_export/session_log.py

Comment thread src/kernelforge/run_export/session_log.py

Comment thread src/kernelforge/run_export/session_log.py Outdated

Comment thread src/kernelforge/run_export/session_log.py Outdated

amd-ama10002-2 force-pushed the feat/kernelforge-run-experiences branch from 05be3a7 to 924313c Compare October 16, 2026 23:34

amd-ama10002-2 force-pushed the feat/kernelforge-export-runs-arena-session branch 2 times, most recently from 811b1c0 to 7e4cf6b Compare October 17, 2026 00:53

amd-ama10002-2 force-pushed the feat/kernelforge-export-runs-arena-session branch 2 times, most recently from 25e84cc to 78558a9 Compare October 17, 2026 02:07

amd-ama10002-2 marked this pull request as ready for review October 17, 2026 02:31

amd-ama10002-2 requested a review from a team as a code owner October 17, 2026 02:31

amd-ama10002-2 and others added 3 commits October 19, 2026 17:28

amd-ama10002-2 force-pushed the feat/kernelforge-run-experiences branch from e5ea4c9 to 225058e Compare October 19, 2026 17:35

amd-ama10002-2 force-pushed the feat/kernelforge-export-runs-arena-session branch from 78558a9 to 132dc38 Compare October 19, 2026 17:36

Navigation Menu

Repository navigation

feat(kernelforge): export-runs reads AgentKernelArena driver.log sessions - #1950

feat(kernelforge): export-runs reads AgentKernelArena driver.log sessions#1950
amd-ama10002-2 wants to merge 3 commits into
feat/kernelforge-run-experiencesfrom
feat/kernelforge-export-runs-arena-session

amd-ama10002-2 commented Oct 16, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Oct 16, 2026 •

edited

Loading

Outside this change (informational): 105

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

amd-ama10002-2 commented Oct 16, 2026

Uh oh!

github-actions Bot commented Oct 16, 2026

Uh oh!

github-actions Bot commented Oct 16, 2026

Uh oh!

amd-ama10002-2 commented Oct 17, 2026 •

edited

Loading

Uh oh!

amd-ama10002-2 commented Oct 17, 2026 •

edited

Loading

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

amd-ama10002-2 commented Oct 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Motivation

What changes

Known limitations

Uh oh!

github-actions Bot commented Oct 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Code metrics gate: PASSED

Outside this change (informational): 105

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟡 Changes recommended

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

amd-ama10002-2 commented Oct 16, 2026

Uh oh!

github-actions Bot commented Oct 16, 2026

Uh oh!

github-actions Bot commented Oct 16, 2026

Uh oh!

amd-ama10002-2 commented Oct 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

PR #1950 -- feat(kernelforge): export-runs reads AgentKernelArena driver.log sessions

Uh oh!

amd-ama10002-2 commented Oct 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

PR #1950 -- feat(kernelforge): export-runs reads AgentKernelArena driver.log sessions

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

amd-ama10002-2 commented Oct 16, 2026 •

edited

Loading

github-actions Bot commented Oct 16, 2026 •

edited

Loading

amd-ama10002-2 commented Oct 17, 2026 •

edited

Loading

amd-ama10002-2 commented Oct 17, 2026 •

edited

Loading