chemenu/instructions/dev/issue-tracking.md

type, name, description

type	name	description
types/instruction.md	issue-tracking	Where open work on this stack is tracked, what the four mandatory area/kind/prio/size labels and the three status flags on a Gitea issue mean, why a status/incoming stub is never implemented as it stands, and how to keep an issue body current across sessions.

Track open work as Gitea issues, not as prose in the repo

Open work on this stack lives at https://gitea.nehmer.net/torben/chemenu/issues, one issue per work package, and nowhere else. There is no TODO.md; there was, and every item in it either became an issue or was already one, described twice.

That is the whole reason for this file: a second list is a second thing to maintain, and the one that drifts is always the one nobody reads first. The issue tracker wins that comparison outright - it has state, comments, labels, and a link that survives the change it describes. A markdown file in the repo has none of it, and it costs a publish to touch.

This instruction exists only in the dev repo. A distributed instance has no issues at that URL, which is exactly why dist export excludes instructions/dev/ wholesale (see tools/CONTRACT.md).

Contents

• When to run
• Steps
• Incoming stubs
• Renames and other decay in the tracker
• Citing an issue in the repo
• What no tool checks
• Decision points

When to run

• Something is worth doing but not now. Open an issue; do not write it down in the repo.
• A session's findings outgrow the change it was making - a gap in the tooling, an assumption nobody has checked, a decision that needs the user.
• Picking an issue up: before doing anything else, read the body as the current spec, and re-label it if the ground has moved since.
• The issue carries status/incoming: it is a human's stub, not a spec, and it is worked out and triaged before anything is built from it (§ Incoming stubs).
• While working on one: the body is updated as the state moves, not at the end (step 2). A session that is interrupted leaves the body as its handover.
• Prioritising: deciding what to pick up next, or re-labelling after the ground moved.
• Closing one: the body is rewritten to its final state first, and only then closed (step 7).
• A rename or move ships: sweep the open issues for text that assumed the old name or path (§ Renames and other decay in the tracker).

Steps

1. Write the issue so it survives without you. What is broken or missing, why it matters, what "done" looks like as acceptance criteria, and the specific files or commands involved. An issue that only makes sense to whoever wrote it is a note, and notes were the problem.

   Destructive steps carry the invariant they must not violate. A body that prescribes a mechanism gets built as prescribed - including its bugs. Where a step deletes, overwrites, resets or moves, name the property that must still hold afterwards, not only the command that gets there. "Remove the working directory, then git checkout HEAD -- <stage>" is a mechanism; "the content stages must afterwards match HEAD exactly, without any untracked or ignored file being touched" is the same instruction plus its test - a build instruction and an acceptance criterion at once, so the defect surfaces while the test is written rather than in review afterwards. #30's upstream merge body wrote the mechanism and got exactly that bug: a working-directory removal that took a stage's gitignored, unrecoverable data with it.

   An acceptance criterion states a checkable property, not an activity. "Implement X" is done when someone says so; "after upstream merge, reports/ still holds every file it held before" is done when it is true. This is not a ban on imperative steps - a numbered procedure can still produce a correct control flow, and that is its merit - it binds the destructive steps, and every box in the criteria list.

2. The body is the working state, not a historical first post - keep it current as you go. It is this stack's plan file: the same thing a harness's own plan document is, and it is maintained the same way. Not written once, not brought up to date at the end, but updated whenever something in it stops being true - a decision made, a criterion met, an approach ruled out, a new constraint found.

   The test is an abort, not a milestone. A session can end at any moment - an interrupt, a context limit, a crash, a human walking away - and whatever the body says at that instant is the entire handover. So the standard is: at every point, a fresh session must be able to open the body and pick the work up from there, without a human re-explaining it and without reading back through the comments. If the body would mislead someone who read it right now, it is already out of date, whether or not the work is finished.

   That means updating during the work, not only at its end:

   • a decision gets made → the decision and its reasoning replace the question
   • an acceptance criterion is done → tick it, in the same session that did it
   • something turns out differently than the issue assumed → the assumption is corrected where it stands, not contradicted three paragraphs later
   • work is deferred or dropped → say so, with the reason, where the criterion is

   Rewrite, never append. Do not add to a text that has become wrong: an additively grown log forces every later reader to reconstruct the current state by filtering the whole history, which is the exact cost the body exists to remove. Comments carry the history (step 3); the body carries the state.

   Body rewrites and comments are an LLM session's job. A human normally touches only labels and metadata directly.

   Reading an issue, the body is the state and comments are history. A session picking an issue up reads the body as the spec; comments are read for provenance - why something was decided, what was tried - never as the current instruction. A recommendation in a comment can be older than the body's decision and read just as convincingly: on #30 an earlier comment recommended a smaller, verify-only command, while the body had since settled on building the full merge command. A session trusting the comment would have built the wrong thing, with a plausible justification out of this repo's own tracker.

   A body that is demonstrably wrong is corrected first, not worked around. "Body beats comment" is a rule of precedence, not a licence to execute a stale spec. Where a comment or the tree proves a claim in the body false, the body is rewritten before the work starts - the rewrite above is the fix; leaning on the comments as the "real" state is not. #10 is the case: its body claimed coverage had never been measured while three comments carried a percentage, a statement count and a CI run number.

   Where two comments contradict each other, evidence decides, not recency. On #10, one comment showed a retrieved artifact with zero items on a finished run - the report was not actually retrievable - and a later comment declared the same criterion met without re-checking. The later comment is not the newer truth, only the unchecked one. Resolve it into the body with the evidence named, or mark the point open.

3. Comment a changelog, never a copy. A body rewrite gets one short comment naming only what changed against the previous state - what is new, what is gone, what was corrected. Do not snapshot the old body into a comment: a full copy per revision forces a human to diff two prose texts, which is not a readable history, only another copy.

   One comment per session's worth of change, not per edit. Step 2 asks the body to be kept current continuously, and a comment for every tick would bury the board in noise; the changelog line summarises what that session moved. Trivial upkeep - a typo, a tightened sentence - needs no comment at all.

   **Changelog:** Decision 2 tightened - `kind/` may now change over an
   issue's life. Old acceptance criterion 3 dropped (covered by #42).
   

4. Give it all four mandatory labels: one area/, one kind/, one prio/, one size/. All four, always. Machine maintenance by an LLM session is what makes four axes affordable - the original objection to a third and fourth axis was the upkeep cost for a single human maintainer, and that objection no longer holds.

   area/	Means
   area/kb	The kb/ schema, contract, provenance machinery, lint - the knowledge base as a system.
   area/distribution	Shipping, upgrading and versioning an instance.
   area/corpus	The content and scope of kb/ in this instance, and the demo/testbed question.
   area/workflow	Git, merging, branching, publish, PRs.
   area/process	The development process itself, rather than the stack as an artefact.

   There is deliberately no area/tools: tooling is filed under the domain it serves, not under where its code sits. The axis follows the stage split in AGENTS.md.

   kind/	Means
   kind/decision	Waiting on an operator decision.
   kind/build	Specified; waiting only on implementation time.
   kind/defect	A finding: documentation and reality, or two documents, contradict each other.

   kind/ is expected to change over an issue's life - decision becomes build once the decision is made. That is session memory working, not a labelling failure.

   prio/	Means
   prio/blocking	Blocks or damages work in progress. Next.
   prio/planned	Accrues interest. Planned.
   prio/waiting	Worth doing, waiting on a trigger.

   prio/waiting is not a graveyard. It means the issue's value is real but gated on something outside it - a decision, another issue, a second instance existing. Name that trigger in the issue, or the label is a polite no.

   size/	Means
   size/S	One session, one publish, a clear cut.
   size/M	Several files; a contract or instruction change; its own test effort.
   size/L	Several sessions, or open design questions before the first commit.

   Size is effort, not importance. A prio/blocking size/S is the best thing on the board; a prio/waiting size/L is a thing to talk about before anyone starts.

5. Add a status/ flag only when it applies. All three are optional, because each describes a temporary condition rather than a property every issue has.

   status/	Means
   status/blocked	Waiting on another, still-open issue - not workable on its own, whatever its prio/ says. Name the blocking issue in the body.
   status/unconfirmed	A reported suspicion, not yet checked against actual behaviour. Applies to any kind/, not just kind/defect.
   status/incoming	A human's stub: a request or a thought, filed at whatever length it arrived, deliberately short of everything step 1 asks for. Never implemented as it stands - § Incoming stubs.

   While status/unconfirmed is set, size/ and prio/ are provisional. Triage ends it one of two ways: the flag comes off and size/prio are set for real, or the issue is closed with the reason. An unverified suspicion does not stay open indefinitely - the process-level analogue of AGENTS.md invariant 3.

   status/incoming is the one flag that suspends step 4 rather than qualifying it. The four mandatory labels are not missing from such an issue, they are not yet due: area/ may be obvious, but kind/, prio/ and size/ are answers to questions the stub has not been read against the tree to settle. Labelling it all four on sight is the failure, not the omission - it makes an unexamined stub look triaged. It is also the one flag a session never adds: an issue a session files meets step 1 or it does not get filed.

6. Re-label when the ground moves, and say why in a comment. A trigger that fired turns prio/waiting into prio/planned. A design question that got answered can drop a size and move kind/decision to kind/build. Silent re-labelling is how a board stops meaning anything.

7. Closing is the last body update, not a comment. If step 2 was followed the body is already nearly there, and closing only settles what the final run established. If it was not, closing is where the whole debt comes due - and it comes due at the worst moment, because a closed body is the version everyone reads afterwards and nobody revisits.

   Either way the body reaches its final state before the issue closes: proposals that were decided read as decided, a "to decide" section has become the decision with its reasoning, acceptance criteria are ticked or struck with a reason, and what was verified is named. Then close, with the one-line changelog comment step 3 asks for.

   Record what actually happened, not a commit hash alone: which proposals were implemented, which were deliberately left out and why, and what was verified. The issue is the only place that record survives - a changelog entry says what changed, not what was decided against.

   A closing report in a comment does not satisfy this. It reads as complete to whoever writes it and leaves a body still phrased as open work: unticked boxes, an undecided decision section, present tense about a defect that no longer exists. #44 closed exactly that way, with a thorough comment above a body that still asked for a decision that had already been made and shipped. Nothing mechanical catches it (see below), which is why it is a step rather than a habit.

Incoming stubs

A status/incoming issue is never implemented as it stands. It is worked out and triaged first, in a session, and only the result of that is built.

The flag exists because the tracker is also the human's inbox, and the two have different entry costs. Step 1 asks for a body that survives without its author - acceptance criteria, files, commands - and a thought worth keeping is not worth that much work at the moment it occurs. So a stub is admitted at whatever quality it arrives, and status/incoming is the receipt: this text was not held to step 1, and nobody should read it as if it had been.

That is the whole danger. A stub looks like a body, and a body is what a session trusts (step 2). What it actually holds is a symptom or a wish - #60 says the confidence defaults "feel too high", #61 says a mechanism from one instruction "would be interesting" elsewhere. Neither states what done means, and the parts they leave out are exactly the parts the human left to be worked out. Building straight from one produces something that matches the sentence, misses the intent, and closes the issue - so the question the stub was standing in for is never asked again. It is step 1's mechanism-versus-invariant lesson one stage earlier: there, a body prescribed a mechanism and got its bugs built; here, a body prescribes nothing at all and gets the gap filled by whoever read it fastest.

Working one out:

1. Read the stub as a statement of intent, not a specification. Its wording is the only evidence of what was actually asked for. Reinterpret it and the record of the request is gone - what remains is the session's reading of it, indistinguishable from the human's.
2. Check it against the tree before rewriting anything. A stub may be a suspicion (status/unconfirmed applies on top where it is), a duplicate of something already built, or a premise that no longer holds. This is the step that decides which of the two exits below the issue takes.
3. Quote the stub verbatim in the elaboration comment, then rewrite the body. Step 2's "rewrite, never append" holds here as everywhere - but the rewrite overwrites the only record of the request, and comments are where history lives (step 3). Here the history is the request.
4. Name the open questions; do not answer them. Where the stub leaves something a session cannot settle from the tree, it stays a question in the body and the issue becomes kind/decision. Guessing turns the human's open question into a spec that reads as decided, which is worse than the stub was: the stub at least announced that it was incomplete.
5. Then step 4 comes due - all four mandatory labels, set against a body that has been read against the tree. That is the moment the stub becomes a work package.
6. Remove status/incoming and leave the one-line changelog comment step 3 asks for.

Triage ends a stub one of two ways, the same two status/unconfirmed has: it is worked out, labelled and the flag comes off, or it is closed with the reason. A stub does not sit in the inbox indefinitely.

Elaboration touches no file in the working tree, so it needs no version bump and no publish - it is tracker work, and several stubs can be worked out in one pass. What comes after it is an ordinary work package, picked up on its merits like any other.

Renames and other decay in the tracker

A rename is not finished when the tree is green. Renaming a package, a path, a command, a flag or the repository itself moves text that lives outside the working tree, and the open issues are the largest such text. Nothing catches them - wikitool does not know this tracker exists and must not learn (see "What no tool checks" below) - so a pass over the open issues is part of the rename, in the session that did it, not a follow-up someone remembers.

Distinguish a wayfinder from a piece of evidence: a path meant to point at where something is gets pulled through; a path quoted for what was true at a time is left standing and dated. Note per corrected body what was pulled through and when, so the next pass can tell a checked body from one that merely looks right. Closed issues are out of scope - they guide nobody.

Renames are not the only thing that ages an issue text. A page a body cites can vanish from kb/ (wikitool search against the cited titles is the second pass), and an old body can carry private infrastructure detail into what is now a public tracker - both found in the same issue, both worth the same look.

Citing an issue in the repo

No file dist export ships may cite an issue number. The board is reachable only from the origin repo, and this very file - the only one that says where it lives - is pruned along with the rest of instructions/dev/. A "#66" that survives into a distributed instance is therefore worse than a dead link: the reader cannot resolve it and cannot tell that it is unresolvable, so a rule appears to rest on evidence nobody can produce. instructions/CONTRACT.md § "Writing an instruction" asks the opposite ("self-contained enough for an agent with no prior context"), and an issue number is the exact counter-example to it.

Which is the same wayfinder/evidence split as in the section below, applied one layer out - but both halves land in the same place here:

• A wayfinder ("see #66 for the reasoning") is resolved: the reasoning goes into the text, and the number goes.
• A piece of evidence ("removed in #66") is dated in words instead - "removed when the schema default was dropped". The sentence carries itself, and the number stays reachable through git blame -> the commit message, which names the issue anyway.

Where a pointer is genuinely worth having here and would leave nothing behind in words, keep it in a <!-- dist:strip-start/end --> block (instructions/CONTRACT.md § instructions/dev/): visible in this repo, removed on export. Two passages use it today.

tools/**/*.py is deliberately outside all of this. A code comment addresses whoever edits that line, and that only ever happens in the origin repo, because dist export prunes the stack-dev skill together with this directory; a distributed tools/ tree is runtime machinery, not reading material. The same holds for .gitignore and tools/.coveragerc - config, not documentation.

docs verify enforces the rule (below).

What no tool checks

wikitool does not know this tracker exists, and should not learn. It ships to instances that have no issues at that URL, while this file and the workflow it describes are pruned by dist export - a Gitea client inside the shipped tool would be a dev-only dependency carried by every instance, to check a board none of them have. The tracker is reachable only through the gitea-mcp server, in a session, by an agent.

So there is no docs verify for the board. Nothing reports a closed issue whose body still reads as open, a body that contradicts its own comments, an issue missing one of the four mandatory labels, or a status/incoming stub that got built as it stood. Every one of those is caught by a session following this file, or not at all - which is the argument for the sequence in step 7 being explicit about the order (body first, then close), rather than leaving it to be inferred from step 2.

The one rule here that is checked is § Citing an issue in the repo, and it is worth being clear about why that is not a contradiction. docs verify's check_no_issue_references compiles #\d+ and reads the text dist_cmd.build_plan() would write. It has no client, no URL and no notion of an issue's state - it cannot tell an open issue from a closed one, or a real number from an invented one. What it knows is that a shipped document is making a reference its reader cannot follow, which is a property of the document, not of the board. That is the line: a check may look at what this repo writes about the tracker; none may look at the tracker.

Decision points

• Issue or changelog? An issue is work that is not done. CHANGES.md is what shipped. A finished change needs both: the entry, and the issue closed with the reasoning.
• Issue or kb/ page? An issue is about this stack and is ephemeral - it closes, and it records a wish. A kb/ page is verified knowledge that stays true. Never put wiki content findings in an issue, and never file a work item as a page.
• Rewrite the body, or add a comment? Rewrite whenever a reader of the body alone would otherwise be misled - a changed decision, a dropped criterion, a new constraint. A comment carries the changelog line for that rewrite, and nothing else that a future session needs in order to act. Closing an issue is always a rewrite - see step 7.
• A status/incoming stub looks trivially implementable? Work it out anyway. "Trivial" is a judgement about the sentence, and the sentence is the part the human wrote down cheaply; what it omits is not visible from it. The elaboration of an obvious stub is short - that is the argument for doing it, not for skipping it.
• An old issue carries only prio/ and size/? Complete it to all four when you touch it, rather than in a sweep. The board reaches the new scheme issue by issue, as each is picked up.