Greptile Summary

This PR enriches the skills truncation notice in applySkillsPromptLimits / resolveWorkspaceSkillPromptState so the agent is told the names of skills that were silently dropped due to the prompt-size budget. A new formatTruncatedSkillNames helper compactly lists omitted skill names (hard-capped at 1 KB via a pre-flight suffix check), and buildSkillsTruncationNote assembles the full enriched message together with guidance to run openclaw skills info <name> or openclaw skills list.

Key changes:

• applySkillsPromptLimits now also returns omittedSkills — correctly computed as params.skills.slice(skillsForPrompt.length), which covers both count-based and char-based truncation paths.
• formatTruncatedSkillNames uses a greedy prefix loop where the suffix (+N more) computed during each iteration exactly represents the final output if the loop stops there, guaranteeing the 1 KB cap is always respected.
• Two new test cases cover name listing for char-budget truncation (existing test extended) and count-limit truncation (new test with a 3-skill workspace and maxSkillsInPrompt: 1).

Confidence Score: 5/5

• This PR is safe to merge — it is a purely additive, well-tested change to the truncation notice with no impact on skill loading or prompt sizing logic.
• The implementation is straightforward: the only new data flowing through the call chain is the list of already-computed omitted skills. The 1 KB cap guarantee is mathematically sound (the pre-flight suffix check ensures every accepted candidate string, including its trailing count, was already verified to fit before being committed). Tests cover both truncation modes. No existing behavior is altered.
• No files require special attention.

_{Last reviewed commit: 4ca82d0}

🔒 Aisle Security Analysis

We found 1 potential security issue(s) in this PR:

const remoteNote = opts?.eligibility?.remote?.note?.trim();
...
prompt: joinPromptSections(params.remoteNote, note, skillsBody),

const labels = macNodes.map((node) => node.displayName ?? node.nodeId).filter(Boolean);
... `Remote macOS node available (${labels.join(", ")}). ...`

function sanitizePromptNote(str: string): string {
  return str.replace(/[\x00-\x1f\x7f]/g, ""); // drop control chars/newlines/tabs
}

const remoteNote = sanitizePromptNote(opts?.eligibility?.remote?.note ?? "").trim();

​// If the prompt is XML-ish, escape:
const safeRemoteNote = remoteNote ? `<remote_note>${escapeXml(remoteNote)}</remote_note>` : "";

Analyzed PR: #42637 at commit 42d1f55

_{Last updated on: 2026-04-28T16:27:41Z}

Codex review: needs real behavior proof before merge. Reviewed July 9, 2026, 5:05 PM ET / 21:05 UTC.

Summary
The PR adds JSON omitted-skill lookup keys to skills truncation notes and tests exact openclaw skills info <name> resolution under tight prompt caps.

PR surface: Source +192, Tests +186. Total +378 across 2 files.

Reproducibility: yes. Source inspection shows current main still emits only a count-based truncation note, and the PR head still has a tight-budget path where fixed-reserve candidate selection can over-truncate before the final notice is rendered.

Review metrics: 1 noteworthy metric.

• Prompt Budget Contract: 1 existing limit behavior changed. The PR changes how maxSkillsPromptChars decides which skills remain in the agent prompt, so tight-budget behavior needs explicit review before merge.

Stored data model
Persistent data-model change detected: vector/embedding metadata: src/skills/loading/compact-format.test.ts. Confirm migration or upgrade compatibility proof before merge.

Root-cause cluster
Relationship: fixed_by_candidate
Canonical: #50677
Summary: This PR is a candidate partial fix for the open skills truncation issue, while the merged compact-description PR solved an adjacent part of the same broad truncation problem.

Members:

• canonical: Skills silently truncated - user-created skills invisible with no warning or prioritization #50677 - The open issue describes skills silently disappearing under prompt truncation and requests surfacing, documenting, or controlling that behavior.
• partial_overlap: fix(skills): preserve compact prompt descriptions #88426 - The merged PR addressed compact prompt descriptions under the same prompt cap but did not add omitted skill lookup keys.
• partial_overlap: Feature Request: Add Skill Priority Configuration #50199 - The open issue asks for skill priority configuration, which overlaps with truncation mitigation but requires a product/config decision beyond this PR.

Proposal only: this assessment does not dispatch repair, suppress jobs, mutate sibling items, close, or merge anything.

Merge readiness
Overall: 🦪 silver shellfish
Proof: 🦪 silver shellfish
Patch quality: 🦪 silver shellfish
Result: blocked until stronger real behavior proof is added.

Overall follows the weaker of proof and patch quality, so missing proof can cap an otherwise strong patch.

Rank-up moves:

• Budget compact candidate selection against the final rendered notice or exact equivalent accounting.
• Refresh the branch against current main and preserve the compact-description prompt-format behavior from the merged skills rewrite.
• Update the PR body with exact-head terminal or live output showing the cap and omitted-key lookup path, with private details redacted.

Proof guidance:

• [P1] Needs stronger real behavior proof before merge: The PR body includes terminal proof, but it targets older head db926508c4; the current head needs exact-head terminal or live output with private details redacted. After adding proof, update the PR body; ClawSweeper should re-review automatically. If it does not, the PR author or someone with repository write access can comment @clawsweeper re-review.

Risk before merge

• [P1] Existing tight skills.limits.maxSkillsPromptChars or per-agent prompt-cap setups can hide more skills than necessary because the branch chooses compact candidates with a fixed reserve instead of the final rendered notice.
• [P1] The skills catalog becomes agent/session context, so over-truncation can make an installed capability invisible for a run or refreshed snapshot.
• [P1] The branch is conflicting with current main and must be refreshed without dropping the merged compact-description prompt-format behavior.
• [P1] The PR body proof is terminal output for older head db926508c4; exact-head proof for 0ff24c7e1157e8d4da94463c10e90a772be11fea is still missing.

Maintainer options:

1. Refresh And Fix Final-Notice Budgeting (recommended)
   Rebase onto current main, keep the compact-description format/version behavior, and compute retained candidates with the exact notice that will be emitted.
2. Accept Conservative Truncation
   Maintainers could accept the fixed reserve, but tight-budget users may lose skills that would fit with the actual shortened notice.
3. Pause For Broader Skills Truncation Work
   If omitted-key recovery should be redesigned with priority or pinning, leave this PR paused and continue in the canonical skills truncation issue.

@clawsweeper automerge

Special instructions:
Rebase or port this PR onto current main; preserve the compact-description prompt format/version behavior from src/skills/loading/workspace.ts; replace the fixed COMPACT_WARNING_OVERHEAD candidate budgeting with exact final rendered notice accounting; add/update focused tests for omitted lookup keys plus compact descriptions under tight maxSkillsPromptChars; keep validation scoped to the skills loading and CLI formatter tests.

Next step before merge

• [P1] A narrow automated repair can port the useful omitted-key behavior onto current main and fix the exact-notice budgeting without a product decision.

Security
Cleared: The diff is limited to skills prompt rendering and tests, JSON-stringifies omitted keys, escapes tag-breaking characters, and does not change dependencies, workflows, lockfiles, or secret handling.

Review findings

• [P1] Budget compact candidates with the final notice — src/skills/loading/workspace.ts:1388-1391

Rebased onto upstream/main (HEAD 3646d37). Addressed Codex review:

• Budgeted omitted-skill guidance inside maxSkillsPromptChars so the emitted prompt stays within the configured cap.
• Advertised identifiers now use the exact lookup key that formatSkillInfo() resolves against, so openclaw skills info <name> is guaranteed to match.
• Added regression test covering tight-cap + many-skills case.
• CHANGELOG entry added.

Verification:

• pnpm test src/agents/skills/compact-format.test.ts src/agents/skills.buildworkspaceskillsnapshot.test.ts src/cli/skills-cli.test.ts - PASS
• pnpm exec oxfmt --check --threads=1 src/agents/skills/workspace.ts src/agents/skills/compact-format.test.ts src/agents/skills.buildworkspaceskillsnapshot.test.ts src/cli/skills-cli.test.ts CHANGELOG.md - PASS
• pnpm exec tsc -p tsconfig.core.json --noEmit - PASS

Ready for another look.

Rebased onto upstream/main; current head is db926508c4.

• Fixed the compact-format.test.ts TS2345 failure by replacing the ambiguous .map(makeEntry) callback usage with explicit single-argument callbacks.
• Added structured ## Real behavior proof content to the PR body with before/after prompt output and exact formatSkillInfo() lookup evidence.
• CI for db926508c4 is green: check, check-test-types, Real behavior proof, and the selected Critical Quality lane all pass.

Local validation:

• pnpm test src/agents/skills — PASS
• pnpm test src/cli/skills-cli.test.ts — PASS
• pnpm exec oxfmt --check --threads=1 src/agents/skills/workspace.ts src/agents/skills/compact-format.test.ts src/agents/skills.buildworkspaceskillsnapshot.test.ts src/cli/skills-cli.test.ts CHANGELOG.md — PASS
• pnpm exec tsc -p tsconfig.core.json --noEmit — PASS
• pnpm check:changed --base upstream/main — PASS

Rebased onto upstream/main; current head is 74883f7955.

• Addressed the P2 review by retrying compact prompt rendering when remoteNote pushes the assembled full-format prompt over maxSkillsPromptChars, before dropping any skills.
• Added a regression test covering a remote eligibility note with a near-budget full-format catalog; all skills remain visible in compact mode and no truncation count is emitted.
• CI for 74883f7955 is green: check, check-test-types, Real behavior proof, and the selected Critical Quality lane pass.

Local validation:

• pnpm test src/agents/skills — PASS
• pnpm exec oxfmt --check --threads=1 src/agents/skills/workspace.ts src/agents/skills/compact-format.test.ts CHANGELOG.md — PASS

ClawSweeper PR egg

🎁 Pass real behavior proof to wake the egg and unlock a hatchable treat.

Heads up: this PR needs to be updated against current main before the new required Dependency Guard check can pass.

Clownfish 🐠 reef update

Thanks for the work here. Clownfish patched this branch directly so the contributor trail stays right where it started.

Source PR: #42637
Validation: pnpm test src/skills/loading/compact-format.test.ts src/skills/loading/skills.test.ts src/cli/skills-cli.test.ts; pnpm check:changed
This keeps attribution in the original current: commits, review context, and changelog notes all stay visible.

fish notes: model gpt-5.5, reasoning medium; reviewed against 0ff24c7.

@imwyvern thanks for the PR. ClawSweeper is still waiting on real behavior proof before this can move forward.

Useful proof can be a screenshot, short video, terminal output, copied live output, linked artifact, or redacted logs that show the changed behavior after the fix. Please redact private tokens, phone numbers, private endpoints, customer data, and anything else sensitive.

Once proof is added to the PR body or a comment, ClawSweeper or a maintainer can re-check it.