Defame1297
38f1ba4e03
Claude Code's (and Copilot's) native plugin installer has zero awareness of .apm/ nesting -- it convention-scans only flat skills/, agents/, commands/, hooks.json at each plugin's root. Confirmed via strings on the installed claude binary and live installs of git@holocron/gitea@holocron/kyberforge@ holocron, all reporting Skills(0) Agents(0) Hooks(0) post ADR-0015's apm conversion. Root cause (apm_cli/core/plugin_manifest.py): apm's plugin.json compiler deliberately strips skills/agents/commands keys, assuming the host already auto-discovers those convention directories -- it has no model of .apm/ being host-visible at all. Separately, apm's own bundle exporter (apm_cli/bundle/plugin_exporter.py, behind `apm pack --format plugin`) implements the correct .apm/ -> flat mapping, but only ever targeted build/<name>-<version>/, a path nothing in marketplace.json's source: points at. scripts/sync-plugin-content.sh wraps that bundle exporter and copies its agents/, skills/, commands/, instructions/, extensions/, and merged hooks.json back into each plugin's own root as a second tracked compiled-output category -- same governance status as .claude-plugin/plugin.json: generated from .apm/, never hand-edited. tests/ subdirectories are excluded from the mirror (dev fixtures, not host-visible runtime content; several hardcode a relative repo-root walk-up sized for the .apm/-nested depth, which breaks when duplicated one level shallower). Applied for real across all 6 plugins and verified two ways: `claude plugin validate --strict` passes on every real plugin directory, and a live `claude --plugin-dir <path> -p "list skills/agents"` behavioral test confirms content is now actually discovered. Also, from the same issue #90 review round: - scripts/check-manifests.sh pointed at each plugin's root-level plugin.json (checking skills/hooks/mcpServers/agents pointer fields) -- that file was a stale near-duplicate of .claude-plugin/plugin.json nothing else read or wrote, now deleted across all 6 plugins. check-manifests.sh is rewritten to validate .claude-plugin/plugin.json instead, and drops the pointer-field checks entirely (nothing to check -- those fields are correctly absent by design). Content-presence drift is now check-plugin-content-sync's job, a new pre-push hook wired in .pre-commit-config.yaml. docs/adr/0017 records the root cause and decision in full, including two rejected alternatives (patching plugin.json's path fields directly -- apm's compiler strips them on every run; pointing marketplace.json at apm pack's build/ output -- a version-suffixed non-source directory nothing can install from without an extra build step). ADR-0015 and CONTEXT.md are updated to point at it. Refs: #90
3.4 KiB
topic, source_keys
| topic | source_keys | ||
|---|---|---|---|
| label-inference |
|
Label inference guide
Maps context-pattern signals from conversation content (an issue being drafted, a bug report, a PR
description) to this repo's Kind/* / Priority/* / Status/* label taxonomy. Used by
gitea-issues and gitea-prs before creating or updating an issue/PR, and directly when the user
asks to label something without naming exact labels.
Scoped labels are mutually exclusive — replace, don't stack
Each of Kind/*, Priority/*, Status/* is treated as a scoped-label group by convention (the /
delimiter naming pattern). Gitea's exclusive flag — the mechanism that would let the server itself
enforce one-label-per-scope — is documented as an org-labels-only setting, and the repo-level
label_write methods used here don't accept it at all. So exclusivity within these scopes is a
convention this skill enforces client-side, not something the server guarantees: applying a new
label within a scope is expected to replace any existing label in that same scope on the target
issue/PR, not add alongside it. When inference
selects a Priority/High label and the issue already carries Priority/Medium, the write should
result in only Priority/High remaining — use replace_labels scoped to that group's labels, or at
minimum remove the superseded label before adding the new one. Never leave two labels from the same
scope applied at once.
Signal → label mapping
Kind/* (what kind of work this is):
| Signal in context | Label |
|---|---|
| Bug report, error, crash, unexpected behavior, "broken", "doesn't work" | Kind/Bug |
| New capability, "add support for", net-new functionality | Kind/Feature |
| Improvement to existing behavior, "make X better", refactor with behavior change | Kind/Enhancement |
| Docs-only change, README/comment/guide updates | Kind/Documentation |
| Vulnerability, credential exposure, injection risk, auth bypass | Kind/Security |
Priority/* (urgency):
| Signal in context | Label |
|---|---|
| "blocking", "critical", "urgent", production-down | Priority/Critical |
| "soon", "high priority", "should do this sprint" | Priority/High |
| No urgency signal present | Priority/Medium (default) |
Status/* (workflow state):
| Signal in context | Label |
|---|---|
| Explicit statement that the work is blocked on something else | Status/Blocked |
Procedure
- Read the conversation context (issue/PR title, body, or the triggering discussion) for the signals above.
- Call
label_read method: "list_repo_labels"(seereferences/labels.md) to get the current label set with IDs — inference must never guess an ID, only a name, then resolve it. - Match inferred label names against the resolved list (case-insensitive). If a scope group already has a different label applied on the target and a new one is inferred for that same scope, plan to replace rather than add (see above).
- Low-confidence inference omits the label. If no signal confidently maps to a
Kind/*value, do not guess — omitKind/*entirely rather than default to one.Priority/Mediumis the one exception: it's the explicit default when no urgency signal is present, not a guess. - Hand the resolved IDs (plus which scopes to replace) to the caller's
issue_write/pull_request_writecall — this skill does not apply labels to an issue or PR itself.