docs: stamp the OpenClaw fixture corpus with its source version - #24
Merged
Merged
Conversation
Nothing recorded which OpenClaw build these frames came from, so the version we actually support was unknowable from the repo. They are from 2026.6.34 (the DNO-950 spike); 2026.7.1-2 is what installs today, so the corpus is already behind. Also records the two capture preconditions that fail silently — allowConversationAccess, and the embedded runtime vs the claude-cli harness — plus the ctx.runId stability that Gram's turn correlation depends on. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TiJy9DrFpsmnD2vEkDHKNv
![cubic-dev-ai[bot]](https://avatars.githubusercontent.com/in/1082092?s=60&v=4){kind=small}
This was referenced Sep 1, 2026
tgmendes
approved these changes
Sep 1, 2026
danielkov
approved these changes
Sep 1, 2026
Thegreatsura
pushed a commit
to Thegreatsura/gram
that referenced
this pull request
Sep 2, 2026
…peakeasy-api#5933) Closes DNO-960. Split from speakeasy-api#5929 so nothing here shares a rollback fate with the schema change. ## Why We ship an OpenClaw observability package but never told anyone how to install it. The generated ZIP README had sections for Claude Code, Cursor, Codex and OpenCode, and nothing for OpenClaw. There was also no runbook for the two things that silently break an OpenClaw install — both of which produce a *partly* working state rather than an obvious failure, which is the worst kind. ## What changed - **Generated package README**: an OpenClaw install section — the directory install, `--force` when replacing, the required `allowConversationAccess` block, the Gateway restart, and the coverage caveat - **`docs/runbooks/openclaw-install.md`**: laptop and server/CI install paths, verification, enforcement semantics, version-pin policy, and the operational gotchas from the DNO-950 spike - **`docs/plugins/overview.md`**: OpenClaw listed alongside the other observability packages and in the `downloadObservabilityPlugin` enum ## The two traps the runbook exists for 1. **`allowConversationAccess`.** Without it the conversation hooks silently never fire — no prompts, no assistant responses, no usage — while tool hooks keep working. A half-configured install looks partly fine. 2. **The harness split.** When a model routes through the Claude CLI harness (`agentRuntime: claude-cli`, which `openclaw models auth login` writes *by default* when a claude-cli profile exists), OpenClaw delegates the model and tool loop out-of-process and its tool/LLM hooks never fire. Those sessions are still covered — by our Claude Code hooks — so this is complementary coverage rather than a hole, but a customer in that mode will otherwise think the integration is broken. ## Notes - `generate.go` is a server implementation file, but the change is purely the customer-facing install text, so it belongs with the runbook rather than the feature PR. - No `hooksGeneratorVersion` bump needed: the README is not part of the rendered hook-plugin subtree that `check-generator-versions` diffs. Confirmed by rendering `export-hook-plugin -published` and checking the file list. - The runbook records that our fixtures were captured from OpenClaw 2026.6.34 while 2026.7.1-2 is what installs today, and documents the qualification procedure for that gap. The fixture-side half is speakeasy-api/agenthooks#24. - Deliberately no nightly canary against OpenClaw `latest`: it would page on OpenClaw's release cadence and registry flakes without telling us what we actually support — the same reasoning that keeps the LiteLLM real-proxy suite on `workflow_dispatch`. ## Related PRs Independent — all three branch from `main`, no stacking: - speakeasy-api#5929 — `feat:` source alias and turn correlation - speakeasy-api#5932 — `mig:` the materialized-view change 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_01TiJy9DrFpsmnD2vEkDHKNv <!-- This is an auto-generated description by cubic. --> --- ## Summary by cubic Adds OpenClaw install docs for the observability plugin — the generated ZIP README now has an OpenClaw section, and a new runbook covers laptop and server/CI installs. Two configuration states silently break coverage without any error, and both are now documented. - The README's OpenClaw section is generated only when a hooks API key exists, so it never describes files that aren't in the ZIP. - The runbook details the two traps: missing `allowConversationAccess` drops prompt and usage capture while tool hooks keep working, and the Claude CLI harness routes sessions through Claude Code hooks, which capture them only if those hooks are also deployed on the machine. - Documents how coverage depends on model-auth mode, the enforcement timeout budgets, the version-pin qualification procedure, and that the hooks binary is fetched from the org's Gram server rather than GitHub. - Adds an OpenClaw package-format section in `docs/plugins/package-format.md`, and lists OpenClaw in `docs/plugins/overview.md`, the `downloadObservabilityPlugin` enum, and the observability slug table in `docs/plugins/publishing.md`. - Closes DNO-960. <sup>Written for commit 024520e. Summary will update on new commits.</sup> <a href="https://cubic.dev/pr/speakeasy-api/gram/pull/5933?utm_source=github" target="_blank" rel="noopener noreferrer" data-no-image-dialog="true"><picture><source media="(prefers-color-scheme: dark)" srcset="https://www.cubic.dev/buttons/review-in-cubic-dark.svg"><source media="(prefers-color-scheme: light)" srcset="https://www.cubic.dev/buttons/review-in-cubic-light.svg"><img alt="Review in cubic" src="https://www.cubic.dev/buttons/review-in-cubic-dark.svg"></picture></a> <!-- End of auto-generated description by cubic. --> --------- Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
![https://www.cubic.dev/buttons/review-in-cubic-dark.svg"><source](https://www.cubic.dev/buttons/review-in-cubic-dark.svg"><source){kind=link}
![https://www.cubic.dev/buttons/review-in-cubic-light.svg"><img](https://www.cubic.dev/buttons/review-in-cubic-light.svg"><img){kind=link}
![https://www.cubic.dev/buttons/review-in-cubic-dark.svg"></picture></a](https://www.cubic.dev/buttons/review-in-cubic-dark.svg"></picture></a){kind=link}
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
3 participants
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.
Part of DNO-960.
Why
Nothing recorded which OpenClaw build the
agenthookstest/fixtures/openclaw/frames were captured from, so the version our decoder is actually tested against was unknowable from the repo.They are from 2026.6.34 (the DNO-950 spike). 2026.7.1-2 is what installs today, so the corpus is already behind — which is exactly the kind of drift that was invisible before.
OpenClaw doesn't version its plugin API independently of OpenClaw itself, so the supported range is whatever we've actually qualified, and that's pinned by this corpus rather than by anything in
codec_openclaw.go.What's here
A README recording:
after_tool_call_blocked.jsonexists separatelyctx.runIdstability that Gram'scanonicalAgentTurnIDdepends on — if a future OpenClaw build breaks it, turn correlation degrades silently rather than erroring, so it's worth checking explicitly when re-recordingallowConversationAccess(without it half the corpus is missing), and the embedded runtime vs the claude-cli harness (under which the tool and LLM hooks never fire at all)Docs only — no code changes.
The customer-facing side of both preconditions is in the Gram repo at
docs/runbooks/openclaw-install.md(speakeasy-api/gram#5929).🤖 Generated with Claude Code
https://claude.ai/code/session_01TiJy9DrFpsmnD2vEkDHKNv
Summary by cubic
Supports DNO-960's version-pin policy by recording the OpenClaw build the
agenthookstest/fixtures/openclaw/corpus was captured from, which was previously unknowable from the repo. The frames come from 2026.6.34; 2026.7.1-2 installs today, so the corpus is already behind — the drift this change makes visible. Docs only; no code changes.allowConversationAccess: trueand the embedded runtime; the claude-cli harness never fires the tool and LLM hooks.ctx.runIdstability across hooks is what Gram's turn correlation depends on, and breaking it degrades silently rather than erroring.Written for commit c9487ef. Summary will update on new commits.