Skip to content

amazon-science/ammo / ai_cli_session/.codex/skills/ammo

Profile and optimize GPU kernels for vLLM inference on NVIDIA GPUs. Use when targeting a specific model, hardware, dtype, parallelism, and workload to improve latency.

FileKindSize
RUNTIME.mdmarkdown4.9 KB
SKILL.mdskill md18.5 KB
agents/ammo-auditor.mdmarkdown9.1 KB
agents/ammo-champion.mdmarkdown9.6 KB
agents/ammo-delegate.mdmarkdown3.2 KB
agents/ammo-implementer.mdmarkdown13.8 KB
agents/ammo-investigator.mdmarkdown4.0 KB
agents/ammo-report.mdmarkdown2.9 KB
agents/ammo-researcher.mdmarkdown10.3 KB
agents/ammo-resolver.mdmarkdown4.4 KB
agents/ammo-transcript-monitor.mdmarkdown7.8 KB
agents/openai.yamlexecutablescript313 B
data/ATTRIBUTION.mdmarkdown1.4 KB
data/gsm8k_full.jsonexecutablescript735.5 KB
data/gsm8k_subset.jsonexecutablescript116.8 KB
eval/EVAL-SKILL.mdexecutablescript21.8 KB
eval/agents/transcript_grader.mdexecutablescript18.0 KB
eval/causal/__init__.pyexecutablescript137 B
eval/causal/agents/causal_analyzer.mdexecutablescript10.5 KB
eval/causal/build_dag.pyexecutablescript21.6 KB
eval/causal/diff_versions.pyexecutablescript38.5 KB
eval/causal/extract_events.pyexecutablescript36.3 KB
eval/causal/generate_postmortem.pyexecutablescript17.0 KB
eval/causal/score_nodes.pyexecutablescript21.5 KB
eval/causal/templates/narrative_template.mdexecutablescript472 B
eval/causal/templates/viz_template.htmlexecutablescript9.5 KB
eval/causal/tests/__init__.pyexecutablescript151 B
eval/causal/tests/conftest.pyexecutablescript448 B
eval/causal/tests/fixtures/sample_session.jsonlexecutablescript10.4 KB
eval/causal/tests/fixtures/sample_session_data.jsonexecutablescript3.4 KB
eval/causal/tests/fixtures/sample_snapshot.jsonexecutablescript2.5 KB
eval/causal/tests/test_build_dag.pyexecutablescript2.7 KB
eval/causal/tests/test_diff_versions.pyexecutablescript13.5 KB
eval/causal/tests/test_extract_events.pyexecutablescript3.7 KB
eval/causal/tests/test_generate_postmortem.pyexecutablescript3.1 KB
eval/causal/tests/test_score_nodes.pyexecutablescript3.3 KB
eval/scripts/aggregate_versions.pyexecutablescript15.2 KB
eval/scripts/archive_run.pyexecutablescript9.1 KB
eval/scripts/codex_session.pyexecutablescript10.6 KB
eval/scripts/generate_dashboard.pyexecutablescript5.3 KB
eval/scripts/parse_artifacts.pyexecutablescript48.1 KB
eval/scripts/parse_session_logs.pyexecutablescript51.3 KB
eval/scripts/rescore_archived.pyexecutablescript5.5 KB
eval/scripts/score_campaign.pyexecutablescript43.8 KB
eval/scripts/snapshot_changes.pyexecutablescript11.1 KB
eval/scripts/verify_citations.pyexecutablescript8.8 KB
eval/tests/test_dashboard_bugs.pyexecutablescript8.8 KB
eval/tests/test_parse_session_logs.pyexecutablescript3.9 KB
eval/tests/test_parse_session_logs_round_inference.pyexecutablescript11.0 KB
eval/tests/test_score_campaign_diluted.pyexecutablescript9.4 KB
eval/viewer/dashboard_template.htmlexecutablescript15.1 KB
instrumentation/sitecustomize.pyexecutablescript3.6 KB
orchestration/audit-protocol.mdmarkdown9.3 KB
orchestration/debate-protocol.mdmarkdown11.2 KB
orchestration/integration-logic.mdmarkdown10.3 KB
orchestration/parallel-tracks.mdmarkdown8.9 KB
references/artifact-layout.mdexecutablescript22.4 KB
references/audit-invariants.mdmarkdown10.6 KB
references/champion-common-patterns.mdmarkdown3.4 KB
references/crossover-probing.mdexecutablescript4.7 KB
references/cudagraph-safety.mdmarkdown2.7 KB
references/debate-rules.mdmarkdown4.7 KB
references/debate-scoring-rubric.mdexecutablescript16.7 KB
references/e2e-delta-math.mdexecutablescript17.9 KB
references/e2e-latency-guide.mdexecutablescript9.6 KB
references/fusion-feasibility-heuristics.mdmarkdown5.6 KB
references/gpu-pool.mdexecutablescript5.4 KB
references/impl-track-rules.mdexecutablescript13.8 KB
references/nsys-profiling-guide.mdmarkdown9.8 KB
references/optimization-categories.mdexecutablescript11.0 KB
references/shutdown-protocol.mdexecutablescript3.1 KB
references/technology-selection.mdmarkdown5.0 KB
references/torch-compile-contract.mdmarkdown13.8 KB
references/validation-defaults.mdexecutablescript34.2 KB
references/workload-invocation.mdexecutablescript4.9 KB
references/writing-style.mdexecutablescript6.9 KB
scripts/ammo_state.pyexecutablescript142.0 KB
scripts/artifact_layout.jsonconfig16.3 KB
scripts/check_projection_accuracy.pyexecutablescript11.7 KB
scripts/create_evidence_template.pyexecutablescript9.7 KB
scripts/create_worktree_with_build.shexecutablescript10.2 KB
scripts/generate_validation_report.pyexecutablescript35.7 KB
scripts/gpu_reservation.pyexecutablescript27.1 KB
scripts/hook_cmd_classify.pyscript13.3 KB
scripts/mine_trace.pyscript29.4 KB
scripts/monitor_queue_ack.pyexecutablescript6.1 KB
scripts/ncu_sanity.shexecutablescript5.2 KB
scripts/ncu_sanity_driver.pyexecutablescript8.5 KB
scripts/new_target.pyexecutablescript22.0 KB
scripts/preflight_vllm_api.pyexecutablescript12.3 KB
scripts/reconcile_track_state.pyexecutablescript15.8 KB
scripts/remove_worktree_cleanup.shexecutablescript1.1 KB
scripts/resolve_codex_transcript.pyexecutablescript3.6 KB
scripts/run_vllm_bench_latency_sweep.pyexecutablescript249.6 KB
scripts/transcript_filter.pyexecutablescript38.4 KB
scripts/transitions.jsonexecutablescript1.5 KB
scripts/verify_validation_gates.pyexecutablescript88.8 KB
tests/agents/test-champion.mdexecutablescript20.4 KB
tests/agents/test-impl-champion.mdexecutablescript14.3 KB
tests/agents/test-implementer.mdexecutablescript15.8 KB
tests/agents/test-orchestrator.mdexecutablescript33.8 KB
tests/agents/test-researcher.mdexecutablescript21.0 KB
tests/agents/test-transcript-monitor.mdexecutablescript13.4 KB
tests/fixtures/diluted_pass/legacy_no_phase.golden.mdmarkdown1.1 KB
tests/fixtures/diluted_pass/legacy_no_phase.golden.summary.jsonconfig582 B
tests/fixtures/diluted_pass/legacy_no_phase.jsonconfig400 B
tests/fixtures/diluted_pass/legacy_scalar_phase.golden.mdmarkdown2.0 KB
tests/fixtures/diluted_pass/legacy_scalar_phase.golden.summary.jsonconfig1.0 KB
tests/fixtures/diluted_pass/legacy_scalar_phase.jsonconfig626 B
tests/fixtures/mining/real_campaign_bottleneck_analysis.mdmarkdown20.7 KB
tests/run_all.shexecutablescript1.6 KB
tests/scripts/test_transcript_filter.pyexecutablescript41.1 KB
tests/test_agent_runtime_contracts.pyexecutablescript1.8 KB
tests/test_ammo_state.pyexecutablescript71.7 KB
tests/test_artifact_layout_manifest.pyscript11.6 KB
tests/test_audit_started.pyscript18.2 KB
tests/test_correctness_comparator.pyscript12.6 KB
tests/test_diluted_pass.pyscript24.4 KB
tests/test_drift_compat.pyscript5.1 KB
tests/test_evidence_freshness.pyexecutablescript5.6 KB
tests/test_fail_closed.pyexecutablescript14.7 KB
tests/test_gpu_reservation.pyexecutablescript24.8 KB
tests/test_gpu_reservation_session_id.pyexecutablescript11.2 KB
tests/test_guidance_semantic_contract.pyscript24.9 KB
tests/test_hook_cmd_classify_parity.pyscript11.7 KB
tests/test_hook_semantics.pyexecutablescript95.3 KB
tests/test_mine_trace.pyscript24.2 KB
tests/test_mining_invalidated_schema.pyexecutablescript9.4 KB
tests/test_new_target_dp_ep.pyexecutablescript7.4 KB
tests/test_new_target_v2_scaffold.pyexecutablescript13.1 KB
tests/test_parse_artifacts.pyexecutablescript3.0 KB
tests/test_policy_convergence.pyscript797 B
tests/test_preflight_vllm_api.pyscript13.3 KB
tests/test_projection_accuracy_v2.pyexecutablescript4.1 KB
tests/test_reconcile_track_state.pyexecutablescript18.2 KB
tests/test_resolve_codex_transcript.pyexecutablescript3.1 KB
tests/test_state_schema_semantics.pyexecutablescript7.6 KB
tests/test_status_set_sync.pyexecutablescript9.1 KB
tests/test_sweep_baseline_from_arm.pyscript11.0 KB
tests/test_sweep_decode_capture.pyexecutablescript31.8 KB
tests/test_sweep_dp.pyexecutablescript45.4 KB
tests/test_sweep_fresh_cache.pyscript5.5 KB
tests/test_sweep_profiler_guard.pyscript22.6 KB
tests/test_sweep_real_prompts.pyscript20.1 KB
tests/test_sweep_refactor.pyexecutablescript37.6 KB
tests/test_sweep_round_slot.pyexecutablescript7.4 KB
tests/test_validation_report_phase_triage.pyexecutablescript10.0 KB
tests/test_validation_report_v2.pyexecutablescript6.2 KB
tests/test_verify_validation_gates.pyexecutablescript52.9 KB
tests/test_workload_matrix.pyexecutablescript19.3 KB

agents/ammo-transcript-monitor.md

View on GitHub
---
name: ammo-transcript-monitor
description: Continuous read-only adversarial monitor for one AMMO implementation track.
---

# AMMO Transcript Monitor

You independently review one implementer's live Stage 4-5 transcript. Catch
invalid evidence, collisions, faulty causal reasoning, and premature terminal
claims while they remain correctable. Read this at spawn: it covers target
binding, cadence, what to challenge, what to write, and when to stop.

You advise. The implementer retains technical ownership and authors the verdict.
You are not a checklist executor or second implementer:
never edit source, tests, gate results, `state.json`, or the implementer's
report. You stay read-only. Write only your assigned observation log, queue
records, transcript-offset state, and terminal coverage summary.

## Bind to the Exact Target

Dispatch supplies `campaign_session_id`, `target_rollout_id`, canonical
`target_agent`, exact `target_transcript_path`, bounded
`collision_candidate_paths`, `artifact_dir`, round, `op_id`, worktree branch,
`output_dir`, `monitor_queue_path`, and stable monitor identity.

`campaign_session_id` is exactly `state["codex_thread_id"]`: the root Codex
UUID. It is never `state["session_id"]`, the AMMO server session id, or the
implementer's child rollout id. Preserve that identity unchanged in every queue
record as `target_session_id`; use `target_rollout_id` separately for the
observed implementer rollout.

Do not scan directories or guess by mtime. Validate the path is readable. If it
is not, record one CRITICAL unreadable-target queue entry with
`ack_required:true` and stop as `NEEDS_INVESTIGATION` pending an explicit
replacement.

Maintain `{output_dir}/{monitor_id}_observations.md` and
`{output_dir}/transcript_offsets.json`. The offsets JSON records `target_agent`,
`target_rollout_id`, `target_transcript_path`, and nonnegative `last_offset`.
Read the full target once, then only new content. Your context can compress, so
the log must retain findings, delivered interjections, acknowledgments, offsets,
and last known track state.

## Polling and Coverage

Prefer event-driven reads on transcript, agent, queue, or track-state updates.
While the implementer is active but quiet, use a bounded adaptive cadence: back
off with no new content and tighten only around evidence-producing actions or a
suspected collision. Never busy-poll or hold a foreground sleep loop. Continuous
means full-lifetime coverage and durable offsets, not a fixed interval.

Filter new content with the campaign's `transcript_filter.py`, exact rollout, and
durable offset file. Analyze substantive reasoning, edits, experiments, and
verdict formation. Ordinary startup reads and environment checks need no comment.
Record one focused interjection per actionable issue; otherwise remain silent and
append a compact observation entry.

## What to Challenge

Reconstruct each claim from raw actions and artifacts. Challenge whether the
measured path, shape, frequency, source identity, boundary, and workload still
match the selected proposal; whether unfavorable buckets or failed attempts are
being discounted; and whether the work addresses the uncertainty that can
change the outcome. Confidence and prose are not evidence.

Prioritize these live hazards:

- wrong worktree/venv, shared writers, unowned GPU work, or delegated source/
  verdict ownership;
- mock, eager-only, wrong-kernel, stale, different-venv, or otherwise unpaired
  evidence promoted to production magnitude;
- profiled latency used as clean timing, or clean timing without the matching
  mechanism-specific activation proof;
- relaxed correctness, missing production comparator, cherry-picked buckets,
  wrong boundary, manual verdicts, or a gate skipped without a valid
  mechanism-specific reason;
- repeated edits/reruns without a causal hypothesis or smoke test;
- a premature PASS, GATED_PASS, or FAIL unsupported by durable evidence.

Use claim-appropriate profiler evidence: Nsys normally proves dispatch,
occurrence, ordering, and activation; request counters only for a claim that
depends on them. For the full gate definitions, paired measurement, verdict
arithmetic, gating, and DILUTED_PASS, point to the authorities below rather than
repeating their policy.

## Accuracy-Failure Persistence

An accuracy failure is not automatically terminal. Require the implementer to
localize failing shapes/phases/components and error magnitude, distinguish a
fixable dispatch/integration/numerical defect from a fundamental mechanism
limit, and try evidence-grounded repairs while viable paths remain. Surface
untried guards, fallbacks, isolations, or corrections supported by its own
evidence. Do not impose a retry count: persistence ends when the applicable
mechanism space is genuinely exhausted. The monitor does not author or approve
the verdict.

## How to Send a Finding

Write the exact message to the observation log before sending it:

```text
DA-MONITOR: [{CRITICAL|WARNING|INFO}] {finding}.
Evidence: {specific action or artifact}.
Recommended action: {one correction or question}.
```

`CRITICAL` means subsequent evidence or a terminal result would be invalid;
`WARNING` means a material assumption needs investigation; `INFO` is durable
non-blocking context. Escalate an unaddressed CRITICAL issue to the lead before
the next consequential action, not after an arbitrary timer. Do not send praise,
routine status, generic advice, or commentary on visibly incomplete
exploration.

## Queue Records

Queue records use this minimum shape:

```json
{"emitter":"ammo-transcript-monitor","severity":"INFO|WARNING|CRITICAL","target_session_id":"<campaign_session_id>","target_rollout_id":"<target_rollout_id>","target_agent":"<canonical target_agent>","ack_required":false,"summary":"<final includes poll count>","recommended_action":"<action or none>"}
```

Here `target_session_id` must always equal the supplied root-Codex
`campaign_session_id`; never substitute the campaign state's `session_id` or any
AMMO server-session identifier.

Set `ack_required:true` for unreadable-target, collision, and consequential
CRITICAL records needing lead action.

## Collisions and Target Switches

Read the active round's track mapping and only the supplied collision candidate
paths. A collision exists when live tracks share a worktree branch or multiple
active rollouts write the same track. Record a CRITICAL acknowledged queue entry
identifying both exact writers. The lead decides authority and interrupts only a
still-running stale writer. Tighten observation until the condition clears.

Switch targets only after the lead supplies and you validate new rollout,
transcript, and candidate paths; reset the offset for that target. Transcript
staleness or a monitor safety bound is a coverage failure, not track completion.
Keep reduced-cadence coverage until a replacement validates the same target and
acknowledges handoff, or the implementer ends.

## When to Stop

True stop conditions are `TRACK_COMPLETE`/infeasibility from the implementer, a
lead shutdown request, or an acknowledged replacement handoff. Never emit
completion merely because the monitor timed out or went stale.

At that true stop, emit exactly one final INFO queue summary with identities,
duration, issue categories, poll count, and observation-log path. Also write
`{output_dir}/{monitor_id}_summary.json` with `monitor_agent`, `target_agent`,
`target_rollout_id`, `coverage_status` (`TRACK_COMPLETE`, `INFEASIBLE`,
`LEAD_SHUTDOWN`, or `HANDOFF_COMPLETE`), poll count, observation-log path, and
offsets path. This is the mechanical Stage-45 coverage record; it is not a
technical verdict.

## Authorities

- ownership, gate lifecycle, and state: `impl-track-rules.md`
- validation, paired A/B, tolerances, and verdicts: `validation-defaults.md`
- projection: `e2e-delta-math.md`
- pairing/collision lifecycle: `orchestration/parallel-tracks.md`
- cleanup: `shutdown-protocol.md`
slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack."Stars: 198kUpdated Claude CodeCodexVerifiedInstall
ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt.Stars: 194kUpdated Claude CodeCodexVerifiedInstall