fix(observe): allow sdk-ts entrypoint in observation hook

[chris-yyau]

Summary

• When running Claude Code via the Agent SDK (e.g. Happy), CLAUDE_CODE_ENTRYPOINT is set to sdk-ts instead of cli
• The Layer 1 entrypoint guard in observe.sh only allowed cli, causing all observations to be silently dropped for SDK users — the continuous learning system was completely non-functional
• This adds sdk-ts to the allowlist since Agent SDK sessions can be human-interactive

Changes

• skills/continuous-learning-v2/hooks/observe.sh: Add sdk-ts to the Layer 1 case entrypoint allowlist (1 line changed)
• Added comment documenting that non-interactive SDK automation is still filtered by the remaining 4 safety layers (ECC_HOOK_PROFILE, ECC_SKIP_OBSERVE, agent_id, path exclusions)

Test plan

• Verified CLAUDE_CODE_ENTRYPOINT=sdk-ts no longer causes observe.sh to exit early
• Verified project-scoped observations are written to ~/.claude/homunculus/projects/<hash>/observations.jsonl
• Verified non-interactive layers (2-5) still filter automated sessions
• Codex review PASSED

🤖 Generated with Claude Code
via Happy

---

Summary by cubic

Fixes dropped observations for Agent SDK sessions by allowing the sdk-ts entrypoint in the observation hook. Restores continuous learning for interactive SDK runs while non-interactive automation stays blocked by later layers.

• Bug Fixes
  • Allow cli|sdk-ts in Layer 1 entrypoint check in skills/continuous-learning-v2/hooks/observe.sh.
  • Add comment noting layers 2–5 (ECC_HOOK_PROFILE, ECC_SKIP_OBSERVE, agent_id, path exclusions) still filter automation.

^{Written for commit 2fb047d. Summary will update on new commits.}

Summary by CodeRabbit

• Bug Fixes
  • Improved handling of SDK TypeScript sessions to allow them to continue processing instead of being prematurely stopped, enabling better support for non-interactive SDK workflows.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 20bb401d-715c-4924-814c-cc32779f5951

📥 Commits

Reviewing files that changed from the base of the PR and between 4bdbf57 and 2fb047d.

📒 Files selected for processing (1)

• skills/continuous-learning-v2/hooks/observe.sh

---

📝 Walkthrough

Walkthrough

Modified entrypoint handling in observe.sh to allow sdk-ts to pass through Layer 1 gating alongside cli, changing the case pattern from "cli) ;;" to "cli|sdk-ts) ;;". Added descriptive comments for sdk-ts handling. This prevents non-human interactive SDK sessions from being filtered out at the initial gating layer.

Changes

Cohort / File(s)	Summary
Entrypoint Gating Logic skills/continuous-learning-v2/hooks/observe.sh	Modified case statement to include sdk-ts alongside cli in Layer 1 entrypoint gating, allowing SDK sessions to proceed past the initial guard. Added contextual comments.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~3 minutes

Possibly related PRs

• fix: harden observe loop prevention #427: Refactors and hardens early entrypoint and session guards in the same observe.sh script, working with the same gating mechanisms.
• fix(observe): 5-layer automated session guard to prevent self-loop observations #399: Modifies entrypoint gating logic in observe.sh to handle how non-cli entrypoints like sdk-ts are processed.
• fix: harden observer hooks and test discovery #513: Hardens and relocates automated session guards in observe.sh while modifying entrypoint pattern matching logic.

Poem

🐰 The gate swings open, sdk-ts walks through,
No longer blocked at Layer One's view,
A rabbit hops past, letting flows renew,
Code gating softens—new paths ring true! ✨

🚥 Pre-merge checks | ✅ 3

✅ Passed checks (3 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and specifically describes the main change: allowing sdk-ts as an entrypoint in the observation hook, which directly matches the changeset's core modification.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

📝 Coding Plan

• Generate coding plan for human review comments

---

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

• X
• Mastodon
• Reddit
• LinkedIn

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

Tip

CodeRabbit can generate a title for your PR based on the changes.

Add @coderabbitai placeholder anywhere in the title of your PR and CodeRabbit will replace it with a title based on the changes in the PR. You can change the placeholder by changing the reviews.auto_title_placeholder setting.

[Copilot]

Pull request overview

This PR prevents continuous-learning observations from being dropped when Claude Code runs via the Agent SDK by allowing the sdk-ts entrypoint to pass the observation hook’s Layer 1 entrypoint guard.

Changes:

• Allow CLAUDE_CODE_ENTRYPOINT=sdk-ts to proceed through Layer 1 in observe.sh.
• Add comments clarifying why SDK sessions are allowed and how non-interactive automation is still filtered by Layers 2–5.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

You can also share your feedback on Copilot code review. Take the survey.

[greptile-apps]

Greptile Summary

This PR fixes a silent observation-drop bug for Claude Code users running via the Agent SDK by adding sdk-ts to the Layer 1 entrypoint allowlist in observe.sh. Previously, any session launched with CLAUDE_CODE_ENTRYPOINT=sdk-ts exited at Layer 1 without recording observations, making the continuous-learning system completely non-functional for those users.

Key changes:

• skills/continuous-learning-v2/hooks/observe.sh: cli|sdk-ts replaces the lone cli in the Layer 1 case pattern (1 line).
• Three explanatory comments are added documenting the rationale and the fact that Layers 2–5 (ECC_HOOK_PROFILE, ECC_SKIP_OBSERVE, agent_id, and path exclusions) still filter non-interactive SDK automation.

Notes:

• The fix is minimal, correct, and well-scoped.
• Layers 2–5 remain unchanged and continue to provide defence-in-depth for automated SDK pipelines.
• Should additional SDK entrypoint variants emerge (e.g. a future sdk-js or sdk-py), this case list will need a corresponding update — a brief inline comment acknowledging this would help future maintainers, but it is not a blocker.

Confidence Score: 5/5

• This PR is safe to merge; the change is a single-line addition to a shell case pattern with no risk of regression.
• The diff is one logical line (adding sdk-ts to an existing allowlist). The remaining safety layers (2–5) are untouched and continue to filter automated SDK sessions. The comments accurately describe the design. No new code paths, no new I/O, no security surface changes relative to what already existed for cli sessions.
• No files require special attention.

Important Files Changed

Filename	Overview
skills/continuous-learning-v2/hooks/observe.sh	Adds sdk-ts to the Layer 1 entrypoint allowlist so Agent SDK sessions can pass through to the remaining safety layers (2–5). The one-line logic change is correct and well-commented; no functional regressions introduced.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["observe.sh triggered (PreToolUse / PostToolUse)"] --> B{"stdin JSON empty?"}
    B -- "yes" --> EXIT0["exit 0 (no-op)"]
    B -- "no" --> C{"disabled marker\nexists?"}
    C -- "yes" --> EXIT0
    C -- "no" --> D{"Layer 1:\nCLAUDE_CODE_ENTRYPOINT"}
    D -- "cli ✅\nsdk-ts ✅ (NEW)" --> E{"Layer 2:\nECC_HOOK_PROFILE=minimal?"}
    D -- "anything else" --> EXIT0
    E -- "yes" --> EXIT0
    E -- "no" --> F{"Layer 3:\nECC_SKIP_OBSERVE=1?"}
    F -- "yes" --> EXIT0
    F -- "no" --> G{"Layer 4:\nagent_id present?"}
    G -- "yes" --> EXIT0
    G -- "no" --> H{"Layer 5:\ncwd matches\nskip paths?"}
    H -- "yes" --> EXIT0
    H -- "no" --> I["Detect project context\n(detect-project.sh)"]
    I --> J["Parse & scrub tool I/O\n(Python)"]
    J --> K["Append to\nobservations.jsonl"]
    K --> L["Lazy-start observer\nif enabled"]
    L --> M["Throttled SIGUSR1\nto observer process"]

Loading

_{Last reviewed commit: "fix(observe): allow ..."}

[cubic-dev-ai]

No issues found across 1 file

---

Since this is your first cubic review, here's how it works:

• cubic automatically reviews your code and comments on bugs and improvements
• Teach cubic by replying to its comments. cubic learns from your replies and gets better over time
• Add one-off context when rerunning by tagging @cubic-dev-ai with guidance or docs links (including llms.txt)
• Ask questions if you need clarification on any suggestion

[leegonzales]

Hit it.

Clean single-line fix with good documentation. The sdk-ts entrypoint should absolutely be allowed through Layer 1 — Agent SDK sessions via Happy and similar tools are human-interactive and deserve observation. The remaining safety layers (2-5) properly filter automation.

One note: this doesn't address the existing observe.sh falls back to legacy output fields when tool_response is null test failure on main — that's an unrelated ENOENT issue in the test harness, not in the entrypoint logic.

Reviewed by Servitor (heartbeat 7).