CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

terminal-bench-loop

Run one Terminal-Bench task through a bounded Paperclip smoke/diagnosis/fix loop. Use when asked to drive Terminal-Bench until it passes, rerun a Terminal-Bench loop, or iterate with board-gated fixes and diagnosis.

Quality

83%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Passed

No findings from the security scan

SKILL.md
Quality
Evals
Security

Quality

Content

70%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

A dense, highly actionable operational runbook with excellent workflow sequencing, comprehensive validation checkpoints, and explicit feedback and stop rules. Its weaknesses are token efficiency (core rules restated verbatim across four to five sections) and progressive disclosure (a 234-line monolith with no reference files despite a clearly separable optional task-tree procedure).

Suggestions

Consolidate the repeated liveness/typed-waiter and dispatch-config restatements (Inputs, Step 3, Step 6, Step 8, Worktree rule, Pitfalls, verification checklist) into the single Liveness rule section and cite it from elsewhere, cutting substantial duplicate tokens.

Move the optional task-tree topology and its 10-step procedure into a reference file (e.g. references/task-tree.md) and keep SKILL.md focused on the default single-issue workflow, since the task tree applies only when explicitly requested.

Add a verbatim example benchmark command block and a sample PAPERCLIP_HARBOR_RUNNER_CONFIG JSON so the smoke setup is copy-paste ready rather than described only by its required keys.

DimensionReasoningScore

Conciseness

Mostly efficient, domain-specific operational content with no generic concept explanations, but key rules are restated nearly verbatim across sections: the typed-waiter enumeration ("execution-policy participant, request_confirmation / ask_user_questions / suggest_tasks interaction, approval, or named human owner") appears four times (issue-topology bullet, Step 6, Step 8, Liveness rule) and the PAPERCLIP_HARBOR_RUNNER_CONFIG dispatch-config requirement is explained in five places (Inputs, Step 3, Worktree rule, Pitfalls, checklist). Not anchor 2 because nothing is filler or explanation of concepts Claude already knows — the redundancy is consolidatable but each restatement carries real operational content.

3 / 5

Actionability

Highly concrete guidance throughout: exact issue titles ("Iteration N: <task-name>"), an idempotency key template ("confirmation:{iterationIssueId}:plan:{revisionId}"), exact API fields (blockedByIssueIds, inheritExecutionWorkspaceFromIssueId, "continuationPolicy to wake_assignee"), named env vars, and the copy-paste smoke command `pnpm smoke:terminal-bench-loop-skill`. Falls short of anchor 5 only because the core benchmark invocation and a sample PAPERCLIP_HARBOR_RUNNER_CONFIG JSON are described by required keys rather than shown as executable examples.

4 / 5

Workflow Clarity

A clear 10-step numbered procedure with explicit validation (the 10-item verification checklist, QA/CTO review gates, the deterministic smoke), feedback loops for error recovery ("If the board posts a superseding comment... open a fresh confirmation", "If the workspace is pruned or the worktree path no longer resolves, the loop is invalid"), and explicit stop rules with exact state transitions. All three anchor-5 features are present, and checkpoints are comprehensive rather than "most", so it does not fit anchor 4.

5 / 5

Progressive Disclosure

Section structure is clear (a dozen well-labeled headers), but the 234-line file has zero bundle references while inlining roughly 150 lines of explicitly optional task-tree topology and procedure ("applies only when the user explicitly requests separate tasks") that would be the natural content of a references/ file, so every invocation loads detail the default workflow never uses. Not anchor 4 because most content is not appropriately placed in the main file; not anchor 2 because structure is substantial and nothing is buried.

3 / 5

Total

15

/

20

Passed

Description

87%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

The description cleanly states both what the skill does and when to use it, with concrete natural trigger phrases and a highly distinctive Terminal-Bench/Paperclip niche. Its only weaknesses are slightly compressed action coverage and omission of a few trigger variants (e.g. "smoke loop", "bench loop") that the body documents.

DimensionReasoningScore

Specificity

"Run one Terminal-Bench task through a bounded Paperclip smoke/diagnosis/fix loop" names the domain plus several concrete actions (smoke, diagnosis, board-gated fix, rerun/iterate), but the action coverage is compressed and omits items the body covers (artifact recording, QA/CTO review), falling just short of anchor 5's comprehensive action list.

4 / 5

Completeness

Both halves are explicit: what ("Run one Terminal-Bench task through a bounded Paperclip smoke/diagnosis/fix loop") and when ("Use when asked to drive Terminal-Bench until it passes, rerun a Terminal-Bench loop, or iterate with board-gated fixes and diagnosis"), matching the anchor-5 pattern of a clear capability statement with concrete trigger phrases.

5 / 5

Trigger Term Quality

"Use when asked to drive Terminal-Bench until it passes, rerun a Terminal-Bench loop, or iterate with board-gated fixes and diagnosis" supplies natural phrases a user would actually say, but it misses common variants the body itself lists ("smoke loop", "bench loop", "loop <task-name> through Paperclip"), keeping it below anchor 5's comprehensive synonym coverage.

4 / 5

Distinctiveness Conflict Risk

The combination of "Terminal-Bench task", "Paperclip smoke/diagnosis/fix loop", and "board-gated fixes" is a clear niche with distinct triggers — a user needing this skill would use these exact terms — so overlap with other skills is minimal.

5 / 5

Total

18

/

20

Passed

Validation

100%

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 16 / 16 Passed

Validation for skill structure

No warnings or errors.

Repository

paperclipai/paperclip

Path

.agents/skills/terminal-bench-loop/SKILL.md

Reviewed

Table of Contents

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.