agentleFS
The-40-ThievesThe-40-Thieves/obsidian-tc/.claude/skills/verify-ticket-premise/SKILL.md

verify-ticket-premise

Skill13 starsChanged 5 days agoApache-2.0View on GitHub

Use before implementing any Linear ticket in this repo — checks the ticket's cited file:line references, counts, and central claim against HEAD and reports HOLDS / PARTLY STALE / FALSE before any code is written.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read the line. Do not infer from the file existing.

    line 36
  2. Very specific:

    Zero hits is a result you must state: Write 0 hits for <exact pattern> .

    line 80
  3. Distinctive:

    Do not build an inventory from one grep: Check the code that *consumes* a symbol, not only its definition.

    line 88

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

verify-ticket-premise

Run this before writing code for a ticket. Every time.

Tickets in this repo go stale fast and stale confidently. A sample from a single session:

ticketclaimedactual
(ticket)"0 of 141 tools populate outputSchema"all 150 did; the fail-closed phase had also shipped
(ticket)"~40% of vector bytes duplicated"0.06% on the real vault — the 40% was a synthetic fixture
(ticket)"blocked on another ticket"that ticket had shipped 5 days earlier
(ticket)runbook step "flip activationRerank"the flag reaches no ranking code — a no-op
(ticket)"path ACL is written twice"identical arg lists; one shared helper, two callers

Acting on any of those as written would have wasted between an afternoon and a week. Two of them would have introduced defects. The check below costs minutes.

Procedure

1. Read the ticket in full, including comments

mcp__claude_ai_Linear__get_issue. The newest comment often supersedes the body — several tickets here carry a RE-SCOPED or CORRECTED header whose body below it is preserved but wrong.

2. Resolve every file:line it cites

For each, distinguish three outcomes — they mean different things:

  • EXACT — the line still holds what the ticket says.
  • MOVED — the symbol exists, at a different line. Harmless; note the new location.
  • GONE — no longer exists. The ticket's reasoning may rest on something deleted.

Read the line. Do not infer from the file existing.

Resolve MOVED-vs-GONE structurally, not with rg. A line number is the most perishable thing a ticket can cite — cli.ts went 1256 → 113 lines in one refactor and five tickets still pointed into the gap. Ask where the symbol is:

just where <Symbol>                      # or: bun run where <Symbol> --path packages/server/src

It reports DECLARED / USED / PROSE-ONLY and exits 1 when nothing declares the symbol. The third bucket is the one rg cannot give you: lines that mention the symbol in a comment or string. In this repo that is not a rounding error — MAX_JUDGED reads as 4 ripgrep lines and 2 real ones, and the two extra are a comment about a constant deleted from another file.

3. Re-derive every number

Tickets quote tool counts, test counts, line counts, coverage, corpus sizes, versions. Check each:

rg -c 'REGISTERED_TOOL_COUNT' packages/server/test/registered-tool-count.ts   # tool surface
wc -l <cited file>                                                            # line counts
npm view <pkg> version                                                        # dependency currency

A number being CI-gated does not make it true of production. embed.dup_ratio 0.400 is a hard gate — and it measures a synthetic 5:1 fixture, not any real vault. Ask what the metric's fixture is before quoting it as evidence.

4. Test the central claim

The one sentence the ticket rests on. Search by capability, not by the ticket's chosen identifier — work frequently ships under a different name than the ticket uses. The ticket asked for a note_quality rollup that had already shipped under exactly that table name; an earlier ticket's fail-closed phase shipped as strictOutputSchema.

5. Report before coding

PREMISE HOLDS         — proceed
PREMISE PARTLY STALE  — proceed with the corrected scope, and fix the body
PREMISE FALSE         — do not implement; close or rewrite the ticket

Rules

  • Zero hits is a result you must state. Write 0 hits for `<exact pattern>` . An empty grep and a real absence are indistinguishable unless you say which you found. just where <Symbol> makes the zero checkable — it exits 1 on "nothing declares this", so an absence claim has an exit code behind it rather than a silent empty result.
  • A comment is not a call site. Prose in this repo cites symbols, tickets and invariants constantly, and a comment written to explain a REMOVAL reads as evidence the thing survives. Two separate wrong verdicts here came from counting prose as code. Check the PROSE-ONLY bucket before quoting any count.
  • Do not build an inventory from one grep. Check the code that consumes a symbol, not only its definition.
  • "Code exists" ≠ "code is reachable." Several subsystems here were built, tested, and wired to nothing. Find a production call site.
  • Prefer closing to implementing. Twice in one session the correct output was a closed ticket and no code.

Applies to more than tickets

The same check applies to any confident claim you did not verify this session — a code comment vouching for an invariant, a doc stating a count, a metric quoted as evidence. Reassurance written next to code ages exactly as badly as documentation and gets read with more trust.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.