BC harbor-f2p-p2p-deep-dive
Deep dive on Harbor trial results for tasks that use SWE-Bench-style F2P (FAIL_TO_PASS) and P2P (PASS_TO_PASS) reference tests. Diagnoses why an agent failed and audits whether a failing task is genuinely hard or unfair (instruction-vs-verifier mismatch). Use when the user asks to analyze, debug, investigate, or deep dive on a Harbor run, trial, or job directory; when the user says "harbor deep dive", "f2p deep dive", or "p2p deep dive"; when a trial scores lower than expected; or when the user wants to assess task fairness, instruction quality, verifier correctness, or whether an F2P or P2P test is reasonable for the instruction given. Works with any Harbor task whose verifier reports F2P/P2P-style results, and with any supported agent (claude-code, opencode, codex). Produces a per-failure root-cause verdict (capability gap / verifier issue / instruction issue) by triangulating verifier output, the agent's implementation, the reference tests, and the instruction.
Deep dive on Harbor trial results for tasks that use SWE-Bench-style F2P (FAILTOPASS) and P2P (PASSTOPASS) reference tests.
As a process C 57/100 · Has gaps — References files that are not bundled: scripts/task_review.py
How to improve
- The SKILL.md body is over 5,000 tokens: move reference detail into references/ and load it when needed.
- The text references files that are not there: add them or drop the references.
- Your own cases (evals/evals.json, 4–6 real requests with expected answers): the full check would then run those instead of a model-drafted suite.
- A spec.yaml with trigger phrases and assertions — a behaviour contract for CI; `skilltest init` writes a template.
Guard findings · 0
✓ No critical or high findings
Files scanned: 2. Evidence is masked. Grey chips explain why severity was lowered.
Against the Agent Skills spec
- warning
body-longSKILL.md body ≈ 5118 tokens (recommended < 5000); move details to references/ - warning
missing-refreference to a missing file: scripts/task_review.py
Process rating: all ten parameters 57/100
- 0Tools and files. 1 referenced file(s) missing: scripts/task_review.py
- 0Result and completion. Does not say what the result is
- 70When it triggers. States when to use, but not when not to
- 70Inputs and preconditions. Inputs and preconditions are listed
- 70Execution cost. Instruction body is 5118 tokens
- 85Steps. 64 steps, 2 vague phrases
- 100Failures and branches. 2 branches, has a failure section
- 100Consistency. Name and required fields are in place
- 100Running it twice. Mutating operations check current state
- 100Progress reporting. Reports progress
- medium Safety rules and hard prohibitions inside a skill: they belong in the system prompt, here they protect nothing
- low The response is described with custom markup (13 tags): a typed call is more reliable
Everything here is measured from the skill text rather than judged by a model, so the numbers are checkable. A parameter weighs more when it is a more common reason for the process to stall.
Quality signals
- +4Description does not say when NOT to use the skill (false activations)
- +3Description length 978: 120–800 characters recommended
- +3Output format is not stated: the model decides each time
- +1No license
- +2Single-language instructions
- +5Description quotes 3 example trigger phrases
- +4Structure: 26 headings
- +3Step-by-step instructions: 64 items
- +4Has examples (9 code blocks)
- +4Reference files are cited in the instructions (1 of 1)
Quality base 70; lint remarks subtract, signals add up to 100. Result: 74.