[JesperDramsch]

A desk run on #514 read "fold the three PyCon Germany series names" as renaming every edition to one name. It also replaced alt_name on 2025 to 2027, and dropped a CfP date when it moved a Python BarCamp Karlsruhe entry from 2024 to 2025. Both were reverted by hand. This PR adds the missing rules to the agent guidance.

AGENTS.md

• "Names": each edition keeps the conference it ran under. alt_name can name the series an edition belongs to (PyMunich 2016 has PyCon Germany). Agents never change an existing conference and never replace or clear an existing alt_name.
• "Complete data": a new paragraph saying the data is a historical record. Filling an empty field is always fine. Replacing a filled-in value needs a primary source showing it is wrong for that edition, with the replaced value as old. No entry is deleted, and no value is dropped when data moves between editions.

critic: three new checks in A. A change to an existing conference, or one that replaces or clears alt_name, fails. A replaced value must be wrong for its own edition. A move to another year must lose no value.

enricher, new-year-checker: one line each that points to the AGENTS.md rules.

desk.md

• Step 5: a request that could mean renaming or merging series ("fold", "merge", "canonicalise") gets a Desk: reply asking how the maintainer wants it, and no worklist item.
• Step 6.3: a change to an existing entry's conference or alt_name is held back with a "rename deferred" note, the same way variations that would rename entries are held back. The ledger (step 8, "Ledger") counts a held-back rename as a deferred outcome.

No code changes. tests/test_desk_guard.py and tests/test_desk_pr_body.py pass (1140 tests).

---

Generated by Claude Code

[JesperDramsch]

Routine review

1. Request. No linked issue; request from the PR body: add the missing agent-guidance rules after the #514 desk run renamed the PyCon Germany editions, replaced their alt_name, and dropped a CfP when it moved Python BarCamp Karlsruhe 2024 to 2025. Instances: AGENTS.md ("Names", "Complete data"), .claude/skills/critic/SKILL.md, .claude/skills/enricher/SKILL.md, .claude/skills/new-year-checker/SKILL.md, .claude/routines/desk.md (steps 5, 6.3.2, 7, ledger).

2. Checked against. Wed Oct 21 15:36:30 UTC 2026 (date). No external facts in this review. The data and branch facts come from the repository: _data/legacy.yml:1371-1373 and commit 7a9be97 on auto/conference-updates.

3. Needs your decision.

1. Should the desk ever apply a rename you approve? As written it can't, so the round never closes. desk.md:72 asks you how you want a rename, and says "Their answer comes back through this step". An answer like "rename X to Y" has none of the trigger words, so it becomes an enricher maintainer-request item. The enricher then keeps conference/alt_name "as they are" (enricher/SKILL.md:23). The critic fails the change (critic/SKILL.md:19). Step 6.3.2 leaves it out "accepted or not" and defers it again (desk.md:87), and step 7 sends it back to you (desk.md:114). Before this PR the critic had no rename ban, so a rename you asked for could get through. Deferred variations already had this loop on the base (desk.md 6.3.3). Recommendation: if you do renames by hand, replace "Their answer comes back through this step" with "the maintainer applies it". If the desk should apply them, exempt a maintainer-request item that answers a linked Desk: question from the conference/alt_name ban in all three places.

4. Findings.

Important

• AGENTS.md:52 and .claude/routines/desk.md:72 block the duplicate removal you asked for in the same 🐍 Conference updates #514 thread. Your comment there (2026-10-21T12:51:55Z) said "Canonicalise PyCon Côte d'Ivoire 2027 drop the ASCII-apostrophe entry". The desk carried it out in 7a9be97 "PyCon Côte d’Ivoire 2027: remove duplicate, titles", which deletes a 13-line entry from _data/conferences.yml, and that commit was kept, not reverted. Under this PR, the same comment runs into two blocks:

  • The word "Canonicalise" makes it a question with no worklist item (desk.md:72).
  • Even answered, "Never delete an entry or drop a value" (AGENTS.md:52) has no exception for a second copy of the same edition.

  The sort does not fold such a pair by itself. merge_duplicates (utils/sort_yaml.py:160-197) keys on the exact conference, and fold_name (utils/tidy_conf/utils.py:59-64) keeps ' and ’ apart. The only other route is a variation that renames the entry, and the new critic/SKILL.md:19 and desk.md:87 block that too. Re-running the importer that created the copy reproduces the case.

  Smallest fix:

  • In AGENTS.md:52, add: "A second entry for the same edition (same series and year) is not history: when the maintainer asks, remove it, and keep its spelling as a titles.yml variation."
  • In desk.md:72, narrow the bullet to renaming or merging series, so a request to drop a duplicate entry stays an item.

Nits: None.

5. Tests. The PR changes documentation only, so it adds no test from the request's invariant. Nothing in the suite checks the new rules (no rename/alt_name/value-loss assertion in tests/test_desk_*.py). Output on the PR head 91afcbe:

• pytest tests/test_desk_guard.py tests/test_desk_pr_body.py -q (uv, Python 3.11, REVIEW.md pins): 1140 passed in 174.62s (0:02:54)
• npm ci: exit 0. npm test: Test Suites: 15 passed, 15 total / Tests: 423 passed, 423 total
• utils/sort_yaml.py --skip_links: exit 0, "Conference data sorting and cleaning completed successfully". git diff --stat -- _data shows archive.yml +30, conferences.yml -30: the sort archiving past editions, not caused by this PR.
• Ruff not run: no Python changed.

6. Bounds.

• The PR was merged at 2026-10-21T15:27:20Z, before this review ran. The findings apply to main, commit 2233359.
• Route: REST for the PR, its diff and 🐍 Conference updates #514. Local clone at refs/pull/515/head; rev-parse matched 91afcbe. The diff was taken against merge-base 33057d7.
• Read in full: the diff and the changed sections of all five files.
• Skimmed: the rest of desk.md, the 🐍 Conference updates #514 body and comments, utils/sort_yaml.py merge logic, and the auto/conference-updates log.
• Not run: the full tests/ suite (no utils/ or tests/ change) and pre-commit.
• REVIEW.md, CONTRIBUTING.md and AGENTS.md were read from origin/main.
• Pre-existing: the deferred-variations flow (desk.md 6.3.3 → step 7 → step 5) already had the same unclosed loop as decision 1.

---

Generated by Claude Code