fix(openclaw): restore 2026.9 transcripts and instruction delivery by chhhee10 · Pull Request #794 · FailproofAI/failproofai

chhhee10 · 2026-09-11T15:52:55Z

What changed

OpenClaw 2026.9.2 moved live transcripts from per-session JSONL files to a per-agent SQLite database. That caused both cloud ingestion and the local FailproofAI dashboard to stop seeing current OpenClaw sessions. This PR also restores model-visible instruct() behavior for OpenClaw PreToolUse policies.

Cloud collection / failproofaid

discovers agents/<agentId>/agent/openclaw-agent.sqlite for every OpenClaw profile
incrementally collects each session’s ordered transcript_events
resets cursor and transform state when OpenClaw rewrites a transcript generation
emits events compatible with the legacy JSONL collector so server-side deduplication remains stable
dynamically discovers new profiles without requiring a daemon restart

FailproofAI local dashboard

restores OpenClaw sessions in the Projects page
reads SQLite-backed sessions in the session viewer using the existing OpenClaw message/tool parser
keeps different OpenClaw profiles in separate projects by (agentId, channel)
uses local when OpenClaw has no channel metadata
merges SQLite sessions with legacy JSONL sessions without showing duplicate rows
exports SQLite event_json records as downloadable JSONL
honors OPENCLAW_STATE_DIR before OPENCLAW_HOME

OpenClaw `instruct()` enforcement

keeps existing organization policies and PreToolUse matchers unchanged
preserves OpenClaw's existing exec → Bash canonicalization
emits a distinct permission: "instruct" verdict from the policy evaluator
delivers the instruction through OpenClaw's model-visible blockReason
temporarily interrupts only the first matching tool attempt
allows retries from the same OpenClaw run and policy for five minutes, preventing an advisory instruction from becoming an infinite block loop
keeps real deny() verdicts permanently blocking
clears retry state when the OpenClaw session ends

The temporary rejection is only the transport used to put the instruction in the model's context. Guardrail activity remains classified as instruct, and transcript/session ingestion is unchanged.

Backward compatibility

Both transcript formats are supported at the same time:

older OpenClaw versions continue using agents/<agentId>/sessions/<sessionId>.jsonl
OpenClaw 2026.9.2+ uses agents/<agentId>/agent/openclaw-agent.sqlite
when both stores contain the same (agentId, sessionId), SQLite wins because it is the live copy

Existing cloud policies do not need source changes. After release, machines only need to upgrade FailproofAI, run failproofai update, and restart each relevant OpenClaw gateway/profile so the packaged plugin is reloaded.

Release

Prepared as stable 1.0.4, including package/Cargo versions and CHANGELOG entries.

npm install -g failproofai@latest
failproofai update
openclaw gateway restart

Validation

cargo fmt --all -- --check
cargo test --workspace
focused local-dashboard suite: 36 passed
OpenClaw policy/evaluator/canonicalization/integration suite: 270 passed
exact Zaum Invoice policy regression: matching Finance escalation returns instruct; unrelated Slack sends remain allowed
bunx tsc --noEmit
bun run lint — 0 errors; 4 existing <img> warnings
bun run test:e2e — 333 passed
clean stable tarball installed in oven/bun:latest; packaged CLI returned permission: "instruct" and packaged retry bridge interrupted once then allowed retry
full unit suite: 5,128 passed, 11 skipped; 7 unrelated local-only failures because this checkout's skip-worktree .codex/hooks.json is absent
read-only smoke test against a real OpenClaw 2026.9.2 database: two SQLite sessions discovered in the local Projects adapter, and one 48-record transcript rendered as 21 dashboard entries

Stack note

This PR currently targets #789 because the SQLite-reader hardening it uses lives on that branch. It can be retargeted to main after #789 merges.

Hermes review

Field	Value
Status	Approved
Reviewed commit	`f190591028c3ea866f5dd4d12d1c2a1c28205b44`
Policy revision	`1d8f31d926828f3bae215c58f5b35baa44acbff0`
Model	`gpt-5.6-terra`
Duration	509s
Updated	2026-09-11T21:50:07.307834705+00:00

Summary

One correctness issue found: ambiguous OpenClaw project slugs can make a valid agent/channel project inaccessible. Focused TypeScript tests and type checking pass in clean containers.

Changes

Adds SQLite-backed OpenClaw transcript collection alongside legacy JSONL.
Adds dashboard discovery, viewing, and export for SQLite sessions.
Adds OpenClaw PreToolUse instruct delivery with retry gating.

Validation

Passed docker run --rm -v /review/input/workspace:/src:ro -w /tmp oven/bun:latest sh -c 'cp -a /src workspace && cd workspace && bun install --frozen-lockfile >/dev/null && bunx vitest run __tests__/hooks/openclaw-instruct-retry-gate.test.ts __tests__/hooks/openclaw-invoice-instruct.test.ts __tests__/lib/openclaw-projects.test.ts __tests__/lib/openclaw-sessions.test.ts __tests__/lib/download-session.test.ts' — Focused OpenClaw policy, project, session, and download tests passed in an isolated clean install. (23s)
Passed docker run --rm -v /review/input/workspace:/src:ro -w /tmp oven/bun:latest sh -c 'cp -a /src workspace && cd workspace && bun install --frozen-lockfile >/dev/null && bunx tsc --noEmit' — TypeScript type check passed in an isolated clean install. (22s)
Skipped docker run --rm --network=none -v /review/input/workspace:/src:ro -w /tmp rust:1.97-bookworm sh -c 'cp -a /src workspace && cd workspace && RUSTUP_TOOLCHAIN=1.97.1-x86_64-unknown-linux-gnu cargo test --offline -p fpai-collect --test openclaw_source' — The isolated Rust container has no cached crates.io dependencies; fetching is unavailable under no-network validation isolation. (1s)

Findings

No blocking findings.

1 advisory finding

Medium/High Project slugs collide for valid agent/channel pairs — openClawProjectName builds openclaw-${agentId}-${channel}. The code explicitly permits hyphens in both values. Thus main-bot/telegram and main/bot-telegram both produce openclaw-main-bot-telegram. getOpenClawProjects emits both folders, but the [name] route resolves by that name and getOpenClawSessionsByEncodedName selects the first matching split, so one project's sessions cannot be opened. (lib/openclaw-projects.ts:141)

Open questions

None.

Policy overrides

None.

github-actions · 2026-09-11T15:53:04Z

Thanks @chhhee10 for your contribution to Failproof AI! 🙌

We'd love to discuss your PR and welcome you to our community.

Discord: https://discord.befailproof.ai/
Reddit: https://www.reddit.com/r/failproofai/

coderabbitai · 2026-09-11T15:53:05Z

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Advanced

Run ID: ca49e77f-723d-4997-ba92-64ccf3450269

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

🔍 Trigger review

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

_{Comment @coderabbitai help to get the list of available commands.}

hermes-exosphere · 2026-09-11T15:54:24Z

Hermes


Status	Reviewing
Verdict	Not reviewed yet
Head	`9a4f12c94da4`
Rounds	0 of 5

No summary yet.

What this changes

No component map for this revision.

Rounds

No review has finished on this pull request yet.

Findings

Nothing raised yet.

@hermes-exosphere help lists every command. This comment is maintained in place — I rewrite it after each review rather than posting a new one.

hermes-exosphere · 2026-09-11T15:59:09Z

Hermes


Status	Reviewed
Verdict	Approved
Head	`f190591028c3`
Rounds	0 of 5

One correctness issue found: ambiguous OpenClaw project slugs can make a valid agent/channel project inaccessible. Focused TypeScript tests and type checking pass in clean containers.

What this changes

flowchart LR
    n0OpenClawSQLitecollector["+ OpenClaw SQLite collector"]
    n1Collectortaskregistration["~ Collector task registration"]
    n2OpenClawpolicybridge["~ OpenClaw policy bridge"]
    n3DashboardSQLitereader["+ Dashboard SQLite reader"]
    n4OpenClawsessionbrowser["~ OpenClaw session browser"]
    n0OpenClawSQLitecollector -- "cursor state and batches" --> n1Collectortaskregistration
    n1Collectortaskregistration -- "roots and collection settings" --> n0OpenClawSQLitecollector
    n2OpenClawpolicybridge -- "instruct verdict and retry scope" --> n2OpenClawpolicybridge
    n3DashboardSQLitereader -- "sessions and event JSON" --> n4OpenClawsessionbrowser

Rounds

Reviewed	Commits in this round	Verdict
`9a4f12c94da4`	`9a4f12c94da4`	Approved
`f52f11bc3f16`	`f52f11bc3f16`	Approved
`f190591028c3`	`f190591028c3`	Approved

Findings

Open

F1 Project slugs collide for valid agent/channel pairs (lib/openclaw-projects.ts) — round 1

@hermes-exosphere help lists every command. This comment is maintained in place — I rewrite it after each review rather than posting a new one.

hermes-exosphere

Hermes found no blocking issues in this revision.

hermes-exosphere

Hermes found no blocking issues in this revision.

hermes-exosphere

Hermes found no blocking issues in this revision.

hermes-exosphere

Hermes found no blocking issues in this revision.

1 advisory finding

Medium/High Project slugs collide for valid agent/channel pairs — openClawProjectName builds openclaw-${agentId}-${channel}. The code explicitly permits hyphens in both values. Thus main-bot/telegram and main/bot-telegram both produce openclaw-main-bot-telegram. getOpenClawProjects emits both folders, but the [name] route resolves by that name and getOpenClawSessionsByEncodedName selects the first matching split, so one project's sessions cannot be opened. (lib/openclaw-projects.ts:141)

fix(openclaw): support SQLite transcripts

9a4f12c

hermes-exosphere approved these changes Sep 11, 2026

View reviewed changes

chore: prepare 1.0.4 stable release

f52f11b

hermes-exosphere approved these changes Sep 11, 2026

View reviewed changes

fix(openclaw): deliver pre-tool instructions

f190591

chhhee10 changed the title ~~fix(openclaw): restore transcripts after SQLite migration~~ fix(openclaw): restore 2026.9 transcripts and instruction delivery Sep 11, 2026

hermes-exosphere approved these changes Sep 11, 2026

View reviewed changes

NiveditJain merged commit 05423cf into feat/audit-secret-leak-scanner Sep 12, 2026
8 checks passed

Navigation Menu

Uh oh!

fix(openclaw): restore 2026.9 transcripts and instruction delivery - #794

fix(openclaw): restore 2026.9 transcripts and instruction delivery#794
NiveditJain merged 3 commits into
feat/audit-secret-leak-scannerfrom
fix/openclaw-2026-9-sqlite

chhhee10 commented Sep 11, 2026 •

edited by hermes-exosphere

Loading

Uh oh!

github-actions Bot commented Sep 11, 2026

Uh oh!

coderabbitai Bot commented Sep 11, 2026 •

edited

Loading

Review skipped

Uh oh!

hermes-exosphere commented Sep 11, 2026

Uh oh!

hermes-exosphere commented Sep 11, 2026 •

edited

Loading

Uh oh!

hermes-exosphere left a comment

Uh oh!

hermes-exosphere left a comment

Uh oh!

hermes-exosphere left a comment

Uh oh!

hermes-exosphere left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

chhhee10 commented Sep 11, 2026 • edited by hermes-exosphere Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What changed

Cloud collection / failproofaid

FailproofAI local dashboard

OpenClaw instruct() enforcement

Backward compatibility

Release

Validation

Stack note

Hermes review

Summary

Changes

Validation

Findings

Open questions

Policy overrides

Uh oh!

github-actions Bot commented Sep 11, 2026

Uh oh!

coderabbitai Bot commented Sep 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Review skipped

Uh oh!

hermes-exosphere commented Sep 11, 2026

Hermes

What this changes

Rounds

Findings

Uh oh!

hermes-exosphere commented Sep 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Hermes

What this changes

Rounds

Findings

Uh oh!

hermes-exosphere left a comment

Choose a reason for hiding this comment

Uh oh!

hermes-exosphere left a comment

Choose a reason for hiding this comment

Uh oh!

hermes-exosphere left a comment

Choose a reason for hiding this comment

Uh oh!

hermes-exosphere left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

chhhee10 commented Sep 11, 2026 •

edited by hermes-exosphere

Loading

OpenClaw `instruct()` enforcement

coderabbitai Bot commented Sep 11, 2026 •

edited

Loading

hermes-exosphere commented Sep 11, 2026 •

edited

Loading