Audit progress signal misleads + invalid-id handling is brittle (feedback from a real open-audit sweep)

[andy-esch]

Feedback gathered while running a full open-audit sweep (triage every open finding into a task, then close fully-triaged audits) against a real planning tree. Version: tskflwctl v0.14.1-0.20260707002012-89afc6655ae2+dirty. Bundled into one issue for context; happy to split into separate ones.

#	Priority	Area	One-liner
P1	High	audit show / audit list	Progress number contradicts its own bar and hides triage state
P2	High	id validation	Non-Crockford id → wrong error, aborts the whole command, no repair path
P3	Medium	audit lint	Legacy finding-status debt has no --fix; was masked by P2
P4	Low	task new --json	created.id is the slug; real minted id only in created.path
P5	Low	lifecycle	No hard-delete for a mistaken/duplicate create

---

P1 — The audit progress signal contradicts itself and hides triage state

The N% in audit show / audit list counts only fixed/landed, but the bar glyphs encode a 3-level gradient (█ fixed, ▓ in-progress, ▒ superseded). So the number and the bar disagree in the same widget:

findings: ▒▒▒▒▒▒▒▒▒▒ 0%  0/5   # all 5 superseded (fully offloaded to tasks) — reads as "0% done"
findings: ▓▓▓▓▓▒▒▒▒▒ 0%  0/4   # 2 in-progress + 2 superseded — bar says "half", number says "0%"
findings: ██████████ 100% 2/2  # all fixed

Consequences:

• An audit that is 100% triaged (0 open, every finding superseded to a real task → correctly closeable) is indistinguishable at the number from an audit nobody has touched — both show 0%.
• This directly caused a false alarm: "0%" was read as "nothing handled / the counter missed findings," when in fact every finding was parsed and superseded.

Suggested fixes (any one helps):

• Label what the number measures (e.g. 0% resolved), or
• Show a stacked count instead of a single % (e.g. 2 superseded · 2 in-progress · 0 fixed), or
• Make the headline number track the same gradient the bar already computes.
• Reconcile with the audit list open column so there's one consistent notion of "progress."

P2 — Invalid-id handling is brittle and misdiagnosed

Some audit files were generated with a non-Crockford char in the id (an l; Crockford base32 excludes i/l/o/u). Result:

1. Wrong error message. not an entity file: "<id>-<date>-<area>.md" has no leading id — move it to meta/ or delete it. The file has a leading id — it's just non-canonical. Correct message: id "<id>" contains non-Crockford-base32 char 'l' (i/l/o/u disallowed).
2. One bad file aborts the whole command. audit list, audit findings, and audit lint print the good rows and then exit 11 — so you can neither trust the exit code nor get clean output, even to query an unrelated audit by slug.
3. No repair path. lint --fix "backfills missing ids" but cannot repair an invalid id — chicken-and-egg: the file isn't discoverable, so the fixer never sees it. (Removing the id + renaming to a bare date-area slug doesn't help either — a no-id filename is hard-rejected the same way.)

The manual fix that worked: substitute the illegal char with its Crockford decode-alias (l→1, i→1, o→0) in both filename and id: frontmatter — same decoded value, canonical char, preserves sort order.

Suggested fixes:

• Precise error naming the offending char / rule.
• Skip-with-warning instead of aborting the whole command (or gate the abort behind --strict).
• lint --fix should repair/re-mint an invalid id, not only backfill a missing one.
• Validate ids at write time so a bad id can never persist (root cause: a generator used a different base32 alphabet than the CLI).

P3 — Legacy finding-status debt has no migration path (and was masked by P2)

• audit lint enforces a strict vocabulary (deferred/fixed/in-progress/landed/open/superseded/wontfix), but older audits use emoji (✅/⏳/⛔) or words (tracked, declined), and the oldest predate the Status: convention entirely.
• lint --fix repairs tasks/epics but there is no audit lint --fix to normalize finding statuses (strip emoji, map declined→wontfix / tracked→superseded, backfill missing).
• The P2 abort hid this: whole-tree audit lint never completed until the invalid-id files were fixed, so ~10 audits' worth of status debt was invisible.
• schema audit documents the Status-line format but not the allowed vocabulary — I had to learn it from lint errors.

Suggested fixes:

• Add audit lint --fix covering the emoji/legacy-word normalizations.
• List the finding-status vocabulary in schema audit.

P4 — task new --json output is confusing

The created object is under created, and created.id is the slug; the real minted id is only embedded in created.path (tasks/<minted-id>-<slug>.md). Scripting created.id expecting the id yields the slug.

Suggested fix:

• Expose the minted id as its own top-level field distinct from slug.

P5 — No hard-delete for a mistaken create

A duplicate task new can only be undone with deprecate (leaves a tombstone) or by rm-ing the file (bypasses the CLI, against the "drive the lifecycle via tskflwctl" principle).

Suggested fix:

• A task rm (and audit rm) for never-committed mistakes.

---

Working well (so this isn't all complaints)

--body-file, --dry-run, multi-slug audit close, schema <kind>, -c column selection, the finding→task supersede model, and the status-grouped tree in audit show are all great. The bar's glyph gradient is a genuinely good idea — it just needs a headline number that agrees with it (P1).