# Audit Career Evidence

> Audit Chronicle project, STAR, and resume claims for evidence quality, ownership, consistency, stale assertions, sensitive information, and unsupported metrics. Use when preparing to write or publish career materials, reviewing captured project records, checking a potentially exaggerated claim, or deciding what evidence must be recovered.

- Skill: `r-jelly/audit-career-evidence` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add r-jelly/audit-career-evidence`
- Raw SKILL.md: https://api.skillmd.com/api/skills/r-jelly/audit-career-evidence/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Security
- Author: r-jelly (https://skillmd.com/u/r-jelly)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/r-jelly/audit-career-evidence

---


# Audit Career Evidence

Assess whether a project story can be defended in a resume or interview. Default
to a read-only report; edit Chronicle files only when the user asks to apply the
audit.

## Scope

Resolve the requested project under
`<vault>/chronicle/projects/<project-id>/`. Read `PROJECT.md`, `TIMELINE.md`,
`OWNERSHIP.md`, `DECISIONS.md`, `EVIDENCE.md`, `STAR.md`, and their referenced
session records when present.

Let `<plugin-root>` be `${CLAUDE_PLUGIN_ROOT}` when Claude Code supplies it;
otherwise use the directory two levels above this `SKILL.md`. Read
`<plugin-root>/docs/DATA_MODEL.ko.md` when a label or evidence rule is unclear.

## Audit Each Claim

Classify every career-relevant claim:

- `confirmed`: direct evidence supports both the action and the attributed
  ownership or result;
- `inferred`: context supports the claim, but a key link is indirect;
- `needs-evidence`: a metric, outcome, ownership statement, or causal claim
  lacks defensible support.

Check:

1. Evidence reference exists and points to the claimed fact.
2. Result claims distinguish observed change from assumed causation.
3. Metrics include unit, baseline, comparison, population, and measurement
   method when those details matter.
4. Personal ownership is distinct from team output.
5. Decision claims include the actual constraints, alternatives, and trade-offs.
6. Project files and session records do not contradict each other.
7. The claim is still current or clearly dated.
8. Public wording does not expose secrets, customer data, internal URLs,
   unreleased plans, or confidential scale figures.
9. Resume language does not strengthen the source claim through words such as
   “led,” “designed,” “owned,” “eliminated,” or “improved” without support.

Never upgrade a claim because it sounds plausible. Never invent a replacement
number.

## Produce the Report

Group findings by severity:

- `Blocker`: unsafe to publish or materially misleading;
- `Needs review`: important ambiguity in ownership, result, or consistency;
- `Evidence opportunity`: useful evidence could be recovered;
- `Ready`: sufficiently supported for the intended use.

For each non-ready claim, provide:

- the exact claim or a short identifying excerpt;
- current and recommended state;
- the gap;
- the smallest useful next action;
- likely evidence sources, without claiming they exist.

End with a compact readiness summary:

```text
confirmed N / reviewed M / audited T
blockers B / needs review R / evidence opportunities E
```

Do not call the audit complete if files or referenced sessions could not be
read.

## Apply Only on Request

If the user asks to apply findings, make targeted edits:

- preserve original wording when adding a state label is enough;
- append missing-evidence tasks to `EVIDENCE.md`;
- record conflicts rather than deleting one side;
- add `user-confirmed:<date>` only after explicit confirmation;
- keep the project or session `captured` until critical review is complete.