learner-history
What it does
learner-history builds a factual, source-verified history of a learner's GitHub issue work across an entire org, and hands back a structured table: what was completed, at what quality, evidenced by what. It never writes a file on its own; it is meant to be called by another skill and read straight out of context.
When to reach for it
Run /learner-history <github-username> before audit-issue or before picking the next issue for that learner, so whichever skill you are running has a real answer to "what has this person already done, and how well." The skill's own description is explicit that this lookup should not be reimplemented inside those callers; it lives here once.
How it resolves each issue
For every issue assigned to the learner in the org, the skill works down a fixed hierarchy of evidence and stops at the first source that gives a clear answer: merged PR branches and commits on the default branch, then GitHub PRs whose branch name or body references the issue, then issue comments (closing notes, partial-work records, admin-close explanations), and finally local session files if any exist. A PR that matched and merged gets full; one that merged alongside others that did not gets partial; a matched but unmerged PR gets in-progress. Where no PR gives a clear answer, the skill falls back to reading the issue's own comments for phrases like "closing on your behalf" or "partial completion" before settling on admin-closed.
Each issue also carries its sequence and tier labels, so the resulting table doubles as a map of where the learner sits in whatever progression the org uses, not just a list of what got closed.
The notes-file pass
Beyond the issue table, the skill looks for per-task notes files (NOTES.md, LEARNING*.md) alongside the task materials, and reports how many exist, how many are actually filled in versus left matching their own prompt text, and whether any have been ingested into a wiki yet. An unfilled section is a concept the learner passed through without articulating it, which is useful signal on its own, and a filled one is raw material: often the first source pages for a learner's own knowledge base.
Capturing a snapshot, deliberately
This is a conversation-only skill by default and stays that way unless you ask otherwise. A history run has a longer shelf life than the conversation that produced it, so at a real milestone (end of a tier, end of a placement, a pay review) it is worth keeping as a dated snapshot. The skill will only raise this if a wiki is already detected, defaults to no when it asks, and never writes into the wiki tree directly: it writes a raw source file to raw/learner-history/<date>-<username>-history.md and tells you to run ingest on it, so that process owns the frontmatter, index, and log updates, and the privacy redaction that go with it.
Because this is a history about a real person, often a junior one, the skill is explicit about what never gets written: no pay figures, no performance judgements, no personal circumstances. Only completion type and evidence source. If the wiki has a placeholder vault for real names, that placeholder is used in place of the learner's own name, and if the learner is a minor, or the wiki is shared beyond the mentor, the skill asks explicitly rather than relying on the default snapshot prompt.
Common questions
Does it ever write to a file by itself? Only the optional dated snapshot, and only after you say yes. Every other run is conversation-only.
What if the learner has more than 100 issues?
The skill pages through with --limit and --skip until it has exhausted the search.
What if there's no git remote to resolve the org from? It asks you directly for the org name instead of guessing.
Where it fits
learner-history is a lookup skill meant to be called from audit-issue or a next-issue-selection flow, and its snapshot output is meant to be picked up by ingest rather than filed directly.