agentsclimarketplace

Log session

Skill jmarek41/surf-guide/.agents/skills/log-session

Log a surf session or visual spot check locally and update private forecast calibration. Use when the user reports what conditions were actually like.From its SKILL.md

Install
npx -y skills add jmarek41/surf-guide --skill log-session

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • 16 days oldThe repository was created 16 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its file declares

Copied from the file, not written here

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

3.0 KB, 608 tokens by cl100k_base, as published. Nobody here has run it

Log a session

Record the user's session report as private ground truth.

Load

Read data/active-location.md first.

If it is missing, tell the user to run /setup-surf.

If its status is none, stop before resolving any region paths. Explain that there is no active destination and offer either:

  • /setup-surf to activate the next trip; or
  • logging against a named archived region, but only after the user explicitly identifies that region.

For an active destination, verify that the region slug and every configured session/calibration path are non-empty. Then read:

  • templates/sessions.csv
  • the active data/sessions/<region>.csv
  • the active data/sessions/<region>.md
  • the active data/calibration/<region>.md

Clarify

Ask only for fields necessary to avoid a misleading record:

  • date, public spot name, and approximate time window;
  • surfed or visual check;
  • observed face height, cleanliness, consistency, wind feel, and verdict;
  • forecast numbers only if they were saved in the prior conversation or the user provides them;
  • a primary and alternate wind-source value when both were captured;
  • what matched and what missed.

Never invent missing forecast values. Leave CSV fields empty.

Write locally

Append:

  1. One concise prose entry to data/sessions/<region>.md.
  2. One CSV row per checked spot to data/sessions/<region>.csv, following the exact template schema.

Update data/calibration/<region>.md only when the observation changes a useful private rule. Preserve counterexamples and evidence counts.

Evaluate a reusable lesson

Follow method/calibration-promotion.md after updating the private record. Classify the result as:

  • personal-only;
  • needs more evidence;
  • candidate public lesson; or
  • update to an existing candidate.

For a candidate, add or update a stable entry in the private calibration's Candidate public lessons table. Keep its evidence references private. Draft a separate anonymized public rule with:

  • public spot name and condition bucket;
  • generalized model or geometry behaviour;
  • observation count and independent-contributor count;
  • counterexamples, uncertainty, evidence label, and confidence;
  • the target locations/<country>/<region>/calibration.md.

Do not edit locations/ during session logging. Show the proposed public rule to the user and ask for explicit approval in a later contribution step.

Privacy

Do not write raw sessions under locations/. If a lesson appears broadly useful, separately offer the anonymized proposal described above.

Preparing a public contribution requires explicit approval. Repeated sessions from this user still count as one independent contributor.

Finish by summarizing what was logged and which private calibration rule, if any, changed. Also report the candidate lesson status without implying that it has already been published.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Keep looking

Skills are one crate of 326,852. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.