From ed7c91c531181ff6ba02c3d941d7996ed4225320 Mon Sep 17 00:00:00 2001 From: Dotta Date: Sat, 3 Oct 2026 09:14:12 -0500 Subject: [PATCH] Record initial bounded ACP Claude dispatch and readiness Co-Authored-By: Paperclip --- ...10-02-stock-harness-paperclip-checklist.md | 8 +++-- .../manifest.json | 30 +++++++++++++++++-- .../2026-10-03-legacy-acp-claude-readiness.md | 6 ++-- 3 files changed, 37 insertions(+), 7 deletions(-) diff --git a/doc/plans/2026-10-02-stock-harness-paperclip-checklist.md b/doc/plans/2026-10-02-stock-harness-paperclip-checklist.md index e38b3b68c27..3d30e0f8c91 100644 --- a/doc/plans/2026-10-02-stock-harness-paperclip-checklist.md +++ b/doc/plans/2026-10-02-stock-harness-paperclip-checklist.md @@ -6,12 +6,12 @@ hire manual is reduced to identity only, and common legacy startup/resume instructions are in [PR #14948](https://github.com/paperclipai/paperclip/pull/14948). GitHub live qualification retains earlier document-delivery failures; the latest focused OpenCode comparison passes both cases in both variants but still exposes -deficient original-case handoff links. That later trial varies only skill sources while holding reduced manuals/shared prompts fixed. Legacy ACP Claude still loses the document only in the original reduced cohort beneath its unchanged credential-guard failure, and no later matched run closes that extra behavioral defect. PR #14948 is CI/review-ready but fails the no-extra-failing-behavior merge criterion and remains draft. The original +deficient original-case handoff links. That later trial varies only skill sources while holding reduced manuals/shared prompts fixed. Legacy ACP Claude still loses the document only in the original reduced cohort beneath its unchanged credential-guard failure, and no later matched run closes that extra behavioral defect. PR #14948 remains draft while the authorized narrow ACP Claude repair is qualified. The new reduced/historical pair is frozen and admitted with zero providers; live outcomes remain pending, and current repair-head CI has one server shard failure under diagnosis. The original [live report](2026-10-02-stock-harness-live-comparison.md), completed [skill repair](2026-10-02-legacy-document-skill-repair.md) and latest [stock-selection/link comparison](2026-10-02-opencode-skill-routing-link-qualification.md) retain their separate sources and limitations. Additional carriers and fixed -native instructions remain open; hiring templates are being reduced separately. +native fixed prompts remain open. Native completion guidance #14961 and hiring templates #14985 are merged; hiring accounting #15007 was externally merged with fail-closed evidence limits preserved. Goal: keep the agent's stock harness behavior and add only what it needs to work with Paperclip. Apply this across legacy adapters, the new Runner, and their @@ -477,6 +477,7 @@ will address them. Record intentional behavior explicitly rather than as a bug. | 2026-10-02 | Both remaining mutable PRs are green after merged-hiring rebase/restack. | #14948 `36aa4d81c`: 52 successful checks / two intentional skips, 599 credential-free prerequisites; #14961 `9138f570c`: 52 successful checks, 637 prerequisites. Both fresh Greptile 5/5, zero unresolved threads, mergeable, draft/unmerged. Prior browser/review failures and all measured refs remain preserved. | | 2026-10-02 | Dotta requires no extra failing behavior before merging, beyond green CI. | PR #14948 does not meet that criterion: legacy ACP Claude loses its document only in the reduced cohort beneath an unchanged credential-guard failure, and subsequent repairs did not select it. Latest OpenCode Pass → Pass is skill-only with reduced manuals held constant, not a historical-manual repeat. Keep #14948 draft; native #14961 has no new failures within its six-pair scope but depends on resolving the base. No new sources, paid runs or merges. | | 2026-10-02 | Bounded hiring notification accounting is ported into both executable count guards on a separate master-based branch. | Draft [PR #15007](https://github.com/paperclipai/paperclip/pull/15007), `778308e032`: Initial 946 support checks and task-only replay are preserved. Current `fed172901` fixes both review findings with 977 support checks / 64 files, typecheck/discovery/contract checks; strict Codex action accounting verifies both variants, ACPX action attribution is unresolved/uncomparable. Original Fail → Fail, source-read gaps and historical body mismatch remain. Exact-head normal CI passes all required gates; fresh Greptile review pending; no models rerun. [Replay](https://github.com/paperclipai/paperclip/blob/fed1729018cc100f5f4bbfb692777e49009c423b/doc/plans/2026-10-02-hiring-executable-accounting-replay.md). | +| 2026-10-03 | Dotta requested #15007 and #14948 ready for merge; no merge delegated here. | #14961 merged at11:20:38Z; #15007 externally merged at11:42:52Z on unchanged fed source after root144 focused checks and exact-head green CI/5of5. #14948 adds shared ACP skill descriptions/staged paths and ephemeral session-env storage. Candidate de096 / historical c256 share current master569c and all common repairs/core fixture/grader/model/auth; exact11 variant paths (2production+9unit). Both actual launcher admissions pass0providers, final96 variant/calibration checks and222 adapter checks pass. One original ACP Claude cell per variant is authorized with enforced single campaign attempts. Initial bounded workflows dispatched once: reduced37127856373/historical37127859382, trusted master569c;0graded/1pending pair. One server workspace provisioning CI assertion and aggregate failure under separate diagnosis. [Manifest and receipts](2026-10-03-legacy-acp-claude-evidence/manifest.json), [readiness report](2026-10-03-legacy-acp-claude-readiness.md). | - F18: Native blocker browser assertion required a marker-only reply despite asking for owner/action/reason. Correct marker-plus-explanation checks symmetrically, calibrate contradictory and future-condition replies, retain original verdicts. - F19: Original prompt-removal cohorts loaded Paperclip; OpenCode skill truncation omitted the late API recipe. The early repair fixed Claude in one paired trial. In the repaired original OpenCode assignment, Paperclip was first loaded only during disposition recovery after a local-file write. Shared legacy operational-skill delivery/selection needs review before more recipe expansion. @@ -512,3 +513,6 @@ Selected #14961 now freezes `d6e59e471` / `e74ed61a6` after the approved five-fi The first actual-provider #14961 comparison is dispatched once on separate immutable targets: [candidate37098728980](https://github.com/paperclipai/paperclip/actions/runs/37098728980) and [historical37098815696](https://github.com/paperclipai/paperclip/actions/runs/37098815696), same trusted workflow215586d. Builds overlap; zero graded pairs at dispatch, usage not yet available. Normal review resolved the publication/baseline approval provenance gates. No paid behavior retry or merge occurred. Selected #14961 first actual cohort now has all twelve original results retained: six candidate passes; historical five passes plus Codex blocker failure for a missing visible marker. Six pairs: zero new failures, one Fail→Pass, five Pass→Pass; no pending pairs. All are single attempts and cleanup passes. [The paired report](2026-10-02-native-completion-master-qualification.md) preserves original failure and exact hashes; independent final audit confirms the historical failure is a missing marker in the actual provider final and identical saved reply, not summary fallback. Its additional native grade/workspace digest was never written after the original browser failure; those missing assertions are not fabricated. The historical Codex completion also has a failed finish call followed by success within the same run (failure reason not retained); all twelve runs/costs and original failures remain retained. Candidate six passes, no extra failing outcomes in six matched pairs, exact-head full CI/fresh 5/5 and zero unresolved threads make #14961 ready for review/merge consideration within this scope; no merge is performed here. Actual invoiced costs and general coding quality remain unmeasured. + + +2026-10-03 current readiness: #14961 and #15007 are merged; their historical report stages above remain intact. The selected #14948 ACP Claude comparison is an initial1+1 current-master repaired-setup cohort, not a reroll of the completed original failure. Operational discovery/env fixes are common; only historical/reduced manual and shared instruction behavior varies. Automatic product recovery stays enabled and counted; campaign behavior retries are disabled. Original credential failure and extra reduced document loss remain preserved until the new paired evidence is available. diff --git a/doc/plans/2026-10-03-legacy-acp-claude-evidence/manifest.json b/doc/plans/2026-10-03-legacy-acp-claude-evidence/manifest.json index db038443063..1b312c766fd 100644 --- a/doc/plans/2026-10-03-legacy-acp-claude-evidence/manifest.json +++ b/doc/plans/2026-10-03-legacy-acp-claude-evidence/manifest.json @@ -1,6 +1,6 @@ { "schema": "paperclip.stock-acp-readiness-comparison.v1", - "status": "admitted-not-dispatched", + "status": "dispatched-initial-pair-zero-graded", "masterContext": "569c7203aa24b95440682983ce7940ba1d4247bd", "variants": { "reduced": { @@ -266,6 +266,30 @@ "7084416945948cb44c9f2070f6356a9c1dece6ec", "f6b56a2bff2ba426213c8b56c9895488402705fe" ], - "liveResults": "pending; no providers launched by local admission", - "claims": "A bounded document delivery and exact-credential safety comparison; no claim of general harness/coding quality." + "liveResults": "0 graded/1 pending pair at dispatch; provider usage/cost not yet available", + "claims": "A bounded document delivery and exact-credential safety comparison; no claim of general harness/coding quality.", + "dispatches": { + "reduced": { + "runId": 37127856373, + "url": "https://github.com/paperclipai/paperclip/actions/runs/37127856373", + "workflowSha": "569c7203aa24b95440682983ce7940ba1d4247bd", + "inputs": { + "target_branch": "codex/stock-acp-reduced-v3-20261003", + "max_parallel": "1", + "all": "false", + "id": "stock-harness.legacy-acp-claude.local.assigned-skill-explicit-invocation" + } + }, + "historical": { + "runId": 37127859382, + "url": "https://github.com/paperclipai/paperclip/actions/runs/37127859382", + "workflowSha": "569c7203aa24b95440682983ce7940ba1d4247bd", + "inputs": { + "target_branch": "codex/stock-acp-historical-20261003", + "max_parallel": "1", + "all": "false", + "id": "stock-harness.legacy-acp-claude.local.assigned-skill-explicit-invocation" + } + } + } } diff --git a/doc/plans/2026-10-03-legacy-acp-claude-readiness.md b/doc/plans/2026-10-03-legacy-acp-claude-readiness.md index 923defb39d6..cd0ffda3439 100644 --- a/doc/plans/2026-10-03-legacy-acp-claude-readiness.md +++ b/doc/plans/2026-10-03-legacy-acp-claude-readiness.md @@ -1,6 +1,6 @@ # Legacy ACP Claude readiness repair -TL;DR: the original reduced cohort has one additional document-delivery defect beneath an unchanged credential failure. Historical ACP Claude saved the requested issue document; reduced ACP Claude did not. Neither ordinary invocation read the staged operational Paperclip skill. The repaired reduced and historical variants have both passed exact-source admission with zero provider calls. One paired ACP Claude case is ready for GitHub; no new live verdict exists yet. +TL;DR: the original reduced cohort has one additional document-delivery defect beneath an unchanged credential failure. Historical ACP Claude saved the requested issue document; reduced ACP Claude did not. Neither ordinary invocation read the staged operational Paperclip skill. The repaired reduced and historical variants have both passed exact-source admission with zero provider calls. The initial paired ACP Claude workflows are running: 0 graded, 1 pending pair. No new live verdict or cost evidence exists yet. ## Preserved observation @@ -25,6 +25,8 @@ Original #14948 head `36aa4d81c49a1a8f6f04b1a068fae19aa901955f` is archived. The Both actual launcher prepare→verify admissions pass with zero providers. See the [immutable manifest](2026-10-03-legacy-acp-claude-evidence/manifest.json), [reduced receipt](2026-10-03-legacy-acp-claude-evidence/reduced-final-preflight.json), and [historical receipt](2026-10-03-legacy-acp-claude-evidence/historical-final-preflight.json). Each prerequisite receipt preserves executed and skipped counts separately; unrelated skipped tests do not count as coverage. The earlier da529/708/f6 zero-provider receipts are superseded and retained locally. The valid review finding now tests historical startup and resumed task/chat carriers separately; startup cannot mask a missing task resume. Complete ordinary task snapshots with an absent conversationMode key normalize to false exactly as production does, while missing/malformed/unscoped snapshots fail closed. -The authorized dispatch is exactly one `stock-harness.legacy-acp-claude.local.assigned-skill-explicit-invocation` cell per variant on separate immutable targets. Each has a 12-minute deadline, company and agent 1,000-cent budget hard stops, enforced `single_attempt` campaign policy, and preserved product recovery with every actual run/cost counted. Core document and exact-credential guards are unchanged. This measures historical versus reduced instructions in the repaired common setup, not a rerun of the old source context or a general coding-quality benchmark. GitHub run IDs and task outcomes remain pending. +The authorized dispatch is exactly one `stock-harness.legacy-acp-claude.local.assigned-skill-explicit-invocation` cell per variant on separate immutable targets. Each has a 12-minute deadline, company and agent 1,000-cent budget hard stops, enforced `single_attempt` campaign policy, and preserved product recovery with every actual run/cost counted. Core document and exact-credential guards are unchanged. This measures historical versus reduced instructions in the repaired common setup, not a rerun of the old source context or a general coding-quality benchmark. The initial workflows were dispatched once, concurrently: [reduced37127856373](https://github.com/paperclipai/paperclip/actions/runs/37127856373) and [historical37127859382](https://github.com/paperclipai/paperclip/actions/runs/37127859382). Both use trusted workflow master569c. At publication, zero cells are graded and the pair is pending; usage/cost are unknown. No behavior rerolls or additional cells. + +Current de096 review completed successfully; the valid instruction-capture thread was fixed/calibrated/resolved. Normal CI has one failed server workspace provisioning test and its aggregate failure, under separate diagnosis. All other normal gates including build/typecheck/Rust/eight browser shards pass. The failing workspace source path is unchanged by the measured ACP/manual work; the observed assertion expected a string PID but received null before its asynchronous persistence. Cause remains under narrow review, not inferred from pass totals. PR #14961 is merged. Parent verified #15007's unchanged `fed1729018cc100f5f4bbfb692777e49009c423b` implementation with 144 focused cases, green CI and fresh 5/5; it was subsequently externally merged. Historical ACPX attribution remains uncomparable; original grades remain intact. #14948 still requires the new matched evidence and current-head checks before readiness.