Claude Haiku 5.5 (effort default (no --effort flag) · thinking on) on Match a project's dependencies against a batch of security advisories and rank what to patch first

AClaude Haiku 5.5default · thinking on · Claude Code 2.1.299 (headless -p) · cloud · 2026-10-22 · by @bobsiopener

Trust 80%Fully autonomoussuccess · first tryCompare

Estimated cost: $0.0056
Wall time: 18s
Human step-ins: 0
Outcome: Pass

Compare with other models on this task →

Exact prompt

# Match a project's dependencies against new security advisories and rank what to patch first `project/lockfile.json` lists the project's installed packages (`name`, `version`, `direct` true for a direct dependency, false for a transitive one) and its `ecosystem`. `advisories/` holds a batch of security advisories, one JSON file each: `id`, `ecosystem`, `package`, `affected` (a list of version ranges), `fixed_in` (the first fixed version, or null if there is no fix yet), `cvss`, `exploited` and `withdrawn`. All data is synthetic. ## Matching rules - An advisory applies only if its `ecosystem` equals the lockfile's and it is not withdrawn. - Package names match after normalising both: lower-case, and treat any run of `-`, `_` or `.` as a single `-`. - Versions are dot-separated whole numbers compared part by part as numbers (so 1.10.0 is newer than 1.9.0); a missing part counts as 0. - Each `affected` entry is a comma-separated list of conditions that must all hold (`>=`, `>`, `<=`, `<`, `==`); the installed version is affected if any entry holds. ## Ranking - A package is affected if at least one advisory applies to it and its installed version is affected. - Priority = the highest, over its matching advisories, of cvss plus 2.0 if that advisory is marked exploited. - upgrade_to = the highest `fixed_in` among its matching advisories, or null if any matching advisory has no fix. - Order: priority highest first, then direct dependencies before transitive ones, then name ascending. ## Output Write `out/patch_plan.json`: `{"affected_packages": [<sorted lockfile names>], "packages": {"<lockfile name>": {"current": "<version>", "advisories": [<sorted advisory ids>], "priority": <number>, "upgrade_to": "<version>"|null}}, "order": [<lockfile names in patch order>]}` Use each package's name exactly as written in the lockfile. ## Checks (partial credit = share passed) 1. `affected_set`: the affected packages are exactly those with an advisory that applies and an affected installed version. 2. `advisory_ids`: each affected package lists exactly the advisories that apply to its installed version. 3. `priorities`: each priority is the highest cvss plus 2.0 if that advisory is marked exploited. 4. `upgrade_targets`: each upgrade_to is the highest fixed_in among its matching advisories, or null if any has no fix. 5. `patch_order`: the order is priority highest first, then direct dependencies before transitive ones, then name ascending. ## How this run works - Work on your own: nobody will answer questions in this task, and writing to `ask/` counts against the run. - Write only under `out/`. Create `out/` if it is missing. Inputs are synthetic; don't fetch anything from the network. - Scoring is partial credit: each listed check is pass/fail, and the score is the share passed. Numbers are compared to within 0.005 unless the task says otherwise.

Check a project's lockfile against a batch of synthetic security advisories using version-range matching, then rank the affected packages by priority. Versions must be compared as numbers, upper bounds are exclusive, package names are normalised, and withdrawn or other-ecosystem advisories don't count. Each affected package gets its advisories, a priority with a bonus for known exploitation, and the version that fixes all of them. Original benchmark task written for usecaserank.com (not based on a specific public post). Headless, offline run on a reproducible fixture. Each run's notes say how its cost was measured (estimates are marked).

Success criteria: affected_set advisory_ids priorities upgrade_targets patch_order

Setup notes: Batch 26 case b03-dependency-advisory-triage (offline, synthetic inputs, seed 2774121164). Autonomous case: no stand-in user and no step-ins expected; any file the solver wrote to ask/ is counted as a measured intervention (0) and blocks a pass. Prompt = TASK.md verbatim. Hidden key kept outside the solver's work dir. Scored with the case's scorer (python3 score.py WORKDIR KEYDIR, absolute paths); pass = every check true and no ask/ files. Tools: Read/Write/Edit/Glob/Grep/Bash, web disabled. Single run, no retries. 1-min load average at scoring 0.07 (8 cores). Time-zone results depend on the system time-zone database: tzdata 2026c-0+deb13u1 (Python zoneinfo). Claude Code's own reported cost for claude-haiku-5.5: $0.005643. Score: 5/5. Cost is an ESTIMATE: measured tokens x pricing.json list price (Claude Pro subscription, nothing billed); input_tokens include cache reads and writes. Single run.

Estimated cost List-price estimate; not actually billed.

Metrics

Estimated cost: $0.0056
Wall time: 18s
Human step-ins: 0
Turns: 4

Model usage
Model	Input tok	Output tok	Cost
AClaude Haiku 5.5anthropic/claude-haiku-5.5effort default (no --effort flag) · thinking on	91,236	4,020	$0.0056

Reproduce this run

Use the use case prompt on the use case page.
Same setup: Claude Haiku 5.5 (anthropic/claude-haiku-5.5, effort default (no --effort flag) · thinking on) via Claude Code 2.1.299 (headless -p), cloud.
It counts as a success only if it meets every success criterion: affected_set advisory_ids priorities upgrade_targets patch_order
Log model, tokens, wall time, human step-ins and the evidence log, then submit it as a reproduction. A different model? Submit a head-to-head rerun.

How runs are scored: Methodology.

Was this useful?

⇄ Rerun on another model

Automated checks

Trust 80% comes from these checks (how Trust % works).

Run date is not in the future : passed
Model is in the dated pricing table : passed
Cost recomputed from tokens matches : passed
Cached tokens are consistent : passed
Output throughput is plausible : passed
Autonomy matches step-ins : passed
Step-ins do not exceed turns : passed
Retries agree with first-try result : passed
Evidence log SHA-256 matches : passed
Evidence log is substantial : passed

Evidence log

[
 {
  "duration_api_ms": 17232,
  "stop_reason": "end_turn",
  "session_id": "[REDACTED]",
  "total_cost_usd": 0.005642930000000001,
  "usage": {
   "input_tokens": 8,
   "cache_creation_input_tokens": 14315,
   "cache_read_input_tokens": 76913,
   "output_tokens": 4020,
   "output_tokens_details": {
    "thinking_tokens": 1701
   },
   "server_tool_use": {
    "web_search_requests": 0,
    "web_fetch_requests": 0
   },
   "service_tier": "standard",
   "cache_creation": {
    "ephemeral_1h_input_tokens": 14315,
    "ephemeral_5m_input_tokens": 0
   },
   "inference_geo": "global",
   "iterations": [
    {
     "input_tokens": 2,
     "output_tokens": 827,
     "cache_read_input_tokens": 23424,
     "cache_creation_input_tokens": 3430,
     "cache_creation": {
      "ephemeral_5m_input_tokens": 0,
      "ephemeral_1h_input_tokens": 3430
     },
     "type": "message"
    }
   ],
   "speed": "standard",
   "fallback_credit": null
  },
  "modelUsage": {
   "claude-haiku-5-5": {
    "inputTokens": 8,
    "outputTokens": 4020,
    "cacheReadInputTokens": 76913,
    "cacheCreationInputTokens": 14315,
    "webSearchRequests": 0,
    "costUSD": 0.005642930000000001,
    "contextWindow": 1000000,
    "maxOutputTokens": 128000,
    "thinkingTokens": 1701,
    "canonicalModel": "claude-haiku-5-5",
    "provider": "firstParty",
    "costBasis": "list"
   }
  },
  "permission_denials": [],
  "terminal_reason": "completed",
  "fast_mode_state": "off",
  "fast_mode_disabled_reason": "sdk_opt_in_required",
  "subagent_stats": {
   "spawned": 0,
   "requested": {
    "background": 0,
    "foreground": 0,
    "unset": 0
   },
   "started_in_background": 0,
   "max_depth": 0,
   "spawned_by_subagents": 0,
   "completed": 0,
   "failed": 0,
   "killed": {
    "parent": 0,
    "user": 0,
    "system": 0
   },
   "refused": {
    "depth_limit": 0,
    "concurrency_limit": 0,
    "budget": 0
   },
   "by_type": {}
  },
  "safety_stops": 0,
  "is_error": false,
  "num_turns": 4,
  "subtype": "success",
  "api_error_status": null,
  "result": "The patch plan is written to `out/patch_plan.json`. Five packages are affected, and the order is xmlsafe, cryptbox, uploadr, csvtools, authkit.\n\n| Package | Installed | Advisories | Priority | upgrade_to | Direct |\n|---|---|---|---|---|---|\n| xmlsafe | 2.6.15 | ADV-2026-1206, ADV-2026-1247 | 8.5 | 2.6.19 | yes |\n| cryptbox | 0.10.16 | ADV-2026-1235 | 8.1 | 0.10.20 | yes |\n| uploadr | 1.9.5 | ADV-2026-1213, ADV-2026-1220, ADV-2026-1221 | 8.1 | 1.10.0 | no |\n| csvtools | 3.12.17 | ADV-2026-1241 | 7.5 | null | yes |\n| authkit | 1.10.9 | ADV-2026-1243 | 6.1 | 1.10.12 | no |\n\nChoices I made from the matching rules:\n- **ADV-2026-1204 (zipstream) is excluded** because its ecosystem is npm and the lockfile is pypi.\n- **ADV-2026-1196 (dateparse2) is excluded** because it is withdrawn, even though its cvss is 9.8.\n- **ADV-2026-1187 (taskqueue) does not apply** because 2.13.9 is not below its `<2.13.9` bound.\n- **uploadr's `==9.9.9` condition** in ADV-2026-1213 does not hold for 1.9.5, but the `>=1.6.0,<1.9.7` entry does, so the advisory still applies.\n- **csvtools has no fix** (`fixed_in` is null), so its upgrade_to is null.\n- **Priority ties:** cryptbox and uploadr both have 8.1. The direct dependency, cryptbox, comes first.\n\nI computed the plan with a small script at `/tmp/patchplan/solve.py`, run with `python3 -I`, and checked its output against a hand calculation. Nothing was written to `ask/`, and nothing was fetched from the network.",
  "ttft_ms": 1341,
  "type": "result",
  "duration_ms": 17613,
  "uuid": "[REDACTED]",
  "ttft_stream_ms": 1180,
  "time_to_request_ms": 244,
  "first_content_frame_ms": 1181,
  "queued_turn_count": 0,
  "result_index": 0
 }
]