)]}'
{".agents/skills/bug-triage/SKILL.md":[{"author":{"_account_id":35345,"name":"Grzegorz Koper","email":"grzegorzk@stackhpc.com","username":"gkoper"},"change_message_id":"1d40b79e9451ebb9ae2bf17cdc60fd46cd2ba98f","unresolved":true,"context_lines":[{"line_number":135,"context_line":"- Check if the bug report references an old version and the described behavior"},{"line_number":136,"context_line":"  is likely to have been fixed since"},{"line_number":137,"context_line":"- If a project checkout is available, search `git log` for commits mentioning"},{"line_number":138,"context_line":"  the bug ID or related keywords, and look for `Closes-Bug` or `Related-Bug`"},{"line_number":139,"context_line":"  references"},{"line_number":140,"context_line":""},{"line_number":141,"context_line":"**RFE (Request for Enhancement)**:"}],"source_content_type":"text/x-markdown","patch_set":1,"id":"fa83f56b_48f6a0a2","line":138,"updated":"2026-05-07 12:58:51.000000000","message":"I think we want to be more explicit here, because in practice the agents misread this.\n\nWhat happened for us was: Hudson comments showed merged patches, and the agent concluded `Fix released`, even though the live Launchpad task was still `New`. Probably because commits used `Related-Bug:` rather than `Closes-bug:`.\n\nIf the intent is \"don\u0027t triage closure from comment text alone\", we would need something more direct here. Like:\n\n\"The live Launchpad task status is authoritative over comment inference, and `Related-Bug:` / Hudson merge comments on their own should not be read as `Fix Committed` or `Fix Released`.\"","commit_id":"12f42d87efbddf8b828cf2761592c4053e74b2d2"},{"author":{"_account_id":35345,"name":"Grzegorz Koper","email":"grzegorzk@stackhpc.com","username":"gkoper"},"change_message_id":"1d40b79e9451ebb9ae2bf17cdc60fd46cd2ba98f","unresolved":true,"context_lines":[{"line_number":178,"context_line":""},{"line_number":179,"context_line":"Present the triage result clearly:"},{"line_number":180,"context_line":""},{"line_number":181,"context_line":"**Validity Assessment**: {category}"},{"line_number":182,"context_line":""},{"line_number":183,"context_line":"**Rationale**: {1-3 sentences citing specific evidence from the bug report,"},{"line_number":184,"context_line":"project knowledge, and source checkout if available}"}],"source_content_type":"text/x-markdown","patch_set":1,"id":"2b532403_46c8ee8a","line":181,"updated":"2026-05-07 12:58:51.000000000","message":"I think we need to be more strict here.\n\nThe classification itself was often fine in our testing; the trouble was that the Launchpad-facing output was still prose-shaped enough that agents could reinterpret it differently later. That shows up much more when agents need to summarise a batch of bugs. \n\nEven a lightweight rule here would help a lot: status must be one exact Launchpad state, importance must be one exact Launchpad importance, and routing should be expressed separately rather than folded into the same line.","commit_id":"12f42d87efbddf8b828cf2761592c4053e74b2d2"},{"author":{"_account_id":35345,"name":"Grzegorz Koper","email":"grzegorzk@stackhpc.com","username":"gkoper"},"change_message_id":"1d40b79e9451ebb9ae2bf17cdc60fd46cd2ba98f","unresolved":true,"context_lines":[{"line_number":187,"context_line":""},{"line_number":188,"context_line":"- Status: {current} → {proposed} (mapping: Configuration Issue → Invalid,"},{"line_number":189,"context_line":"  Unsupported Feature → Won\u0027t Fix, Incomplete Report → Incomplete, Not"},{"line_number":190,"context_line":"  Reproducible in Master → Invalid, RFE → Invalid, Likely Valid Bug →"},{"line_number":191,"context_line":"  Triaged/Confirmed)"},{"line_number":192,"context_line":"- Importance: {current} → {proposed} (for RFE: Wishlist; for Likely Valid Bug:"},{"line_number":193,"context_line":"  proposed level; others: unchanged)"}],"source_content_type":"text/x-markdown","patch_set":1,"id":"95881879_fee449d8","line":190,"updated":"2026-05-07 12:58:51.000000000","message":"Agents also get confused at `RFE -\u003e Invalid`.\n\nModels tent to read \"this is not realy a bug, it is a feature requests\" and jump straight into `Invalid`. That is understandable from the category wording, but does not fit well for Launchpad, where `Wishlist` is a thing, and a lot of projects still keep RFEs opened.\n\nI think this wants tighter wording so bots can’t misread “not a bug” as “must be Invalid”. Even if project policy varies a bit, I would avoid presenting `Invalid` as the obvious default outcome here.","commit_id":"12f42d87efbddf8b828cf2761592c4053e74b2d2"},{"author":{"_account_id":35345,"name":"Grzegorz Koper","email":"grzegorzk@stackhpc.com","username":"gkoper"},"change_message_id":"1d40b79e9451ebb9ae2bf17cdc60fd46cd2ba98f","unresolved":true,"context_lines":[{"line_number":188,"context_line":"- Status: {current} → {proposed} (mapping: Configuration Issue → Invalid,"},{"line_number":189,"context_line":"  Unsupported Feature → Won\u0027t Fix, Incomplete Report → Incomplete, Not"},{"line_number":190,"context_line":"  Reproducible in Master → Invalid, RFE → Invalid, Likely Valid Bug →"},{"line_number":191,"context_line":"  Triaged/Confirmed)"},{"line_number":192,"context_line":"- Importance: {current} → {proposed} (for RFE: Wishlist; for Likely Valid Bug:"},{"line_number":193,"context_line":"  proposed level; others: unchanged)"},{"line_number":194,"context_line":""}],"source_content_type":"text/x-markdown","patch_set":1,"id":"a9c83513_de8d515b","line":191,"updated":"2026-05-07 12:58:51.000000000","message":"`Triaged/Confirmed` is another place where our bots got a bit too loose in my opinion.\n\nModels read both as acceptable and then picks one somewhat opportunistically. In practise we had numerous runs that would classify the same report differently.\n\nI’d suggest tightening this so the automation can’t treat the two states as interchangeable. For example: default to `Triaged`; use `Confirmed` only when there is clear independent corroboration or reproduction ?","commit_id":"12f42d87efbddf8b828cf2761592c4053e74b2d2"}],"/PATCHSET_LEVEL":[{"author":{"_account_id":35345,"name":"Grzegorz Koper","email":"grzegorzk@stackhpc.com","username":"gkoper"},"change_message_id":"1d40b79e9451ebb9ae2bf17cdc60fd46cd2ba98f","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":1,"id":"c9ef0c86_c1aa8e4e","updated":"2026-05-07 12:58:51.000000000","message":"Hi, nice skill! Got some hopefully useful feedback after testing the skill against a existing Kayobe and Kolla-ansible Lauchpads\nI\u0027ve used codex and claude CLIs (with gpt5.5 and opus4.7) to triage 20 new/undecided bugs. \n\nTriagers were mostly on the same page, which is a good sign ;)\nFew places they disagreed on could be fixed by, i believe, making bots follow Launchpad semantics more strictly","commit_id":"12f42d87efbddf8b828cf2761592c4053e74b2d2"}]}