summaryrefslogtreecommitdiffhomepage
path: root/GLOSSARY.md
diff options
context:
space:
mode:
authorAdam Malczewski <[email protected]>2026-06-10 17:03:23 +0900
committerAdam Malczewski <[email protected]>2026-06-10 17:03:23 +0900
commitf6b45507210e04e9884256b0132900640de4334b (patch)
tree73c5779bf2eec5a1d03732be0f6a8b698b8a2c7f /GLOSSARY.md
parentbf862168f0fd7b10d02ae04a9d82f7c37b9d85e5 (diff)
downloaddispatch-f6b45507210e04e9884256b0132900640de4334b.tar.gz
dispatch-f6b45507210e04e9884256b0132900640de4334b.zip
feat(skills): skill system + load_skill tool via per-turn tools filter
Skills are markdown in .skills/ dirs (~/.skills + <cwd>/.skills, cwd shadows home; name = filename). Format: line1 summary, line2 ---, body line3+; load strips the first two lines; malformed = no summary but still loadable. Mechanism (first use of the context-assembly filter chain, §3.2): - kernel: expose HostAPI.applyFilters (delegates to bus.applyFilters) - session-orchestrator: define/export toolsFilter + ToolAssembly; apply once per turn before runTurn (cache-stable across steps), threading cwd + conversationId - skills (new ext): pure parse/merge/render + load_skill tool (live read, path-contained) + a toolsFilter filter rewriting load_skill's description + name enum per cwd - host-bin: register skills in CORE_EXTENSIONS - transport-http: fix HostAPI test stub for the new applyFilters method (fan-out) 734 vitest + 109 bun = 843 tests; tsc -b EXIT 0; biome clean; clean live boot.
Diffstat (limited to 'GLOSSARY.md')
-rw-r--r--GLOSSARY.md3
1 files changed, 3 insertions, 0 deletions
diff --git a/GLOSSARY.md b/GLOSSARY.md
index 39e618f..259dc27 100644
--- a/GLOSSARY.md
+++ b/GLOSSARY.md
@@ -43,6 +43,9 @@
| **key** | The API key (the secret string) held by a credential. | apiKey / api-key (when meaning the whole credential profile) |
| **model name** | The selectable identifier in `<credentialName>/<model>` form — what the model catalog lists and what the CLI / `/chat` `model` field take. | model reference, model id |
| **model catalog** | The list of available model names; served by `GET /models`, aggregated per credential from each provider's `listModels()`. | model list |
+| **skill** | A reusable instruction document (markdown) under a `.skills/` directory, loaded on demand into the conversation by the `load_skill` tool. Discovered from `~/.skills` (home) and `<cwd>/.skills` (project); on a name clash the cwd skill shadows the home one. A skill's name is its filename without `.md`. | prompt snippet, macro |
+| **skill summary** | A skill file's "when to use this skill" line: line 1 of the md, valid only when line 2 is exactly `---`. Advertised (per-turn, cwd-aware) in the `load_skill` tool's description; on load the first two lines are stripped. A file lacking the `---` delimiter shows no summary but stays loadable. | — |
+| **tools filter** | The per-turn `FilterDescriptor` (`toolsFilter`, owned by `session-orchestrator`) through which extensions transform a turn's tool set before it reaches `runTurn`. Applied ONCE per turn so the tool definitions stay byte-stable across steps (prompt-cache safe). The first concrete use of the §3.2 context-assembly filter chain. | — |
## Known vocabulary drift