@counterposition/skill-pi 0.74.0 → 0.74.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/package.json
CHANGED
|
@@ -49,7 +49,8 @@ Use `/logout` to clear stored OAuth credentials. Pi 0.71.0 removed built-in Goog
|
|
|
49
49
|
{
|
|
50
50
|
"anthropic": { "type": "api_key", "key": "sk-ant-..." },
|
|
51
51
|
"openai": { "type": "api_key", "key": "!op read 'OpenAI API Key'" },
|
|
52
|
-
"google": { "type": "api_key", "key": "GEMINI_API_KEY" }
|
|
52
|
+
"google": { "type": "api_key", "key": "GEMINI_API_KEY" },
|
|
53
|
+
"together": { "type": "api_key", "key": "TOGETHER_API_KEY" }
|
|
53
54
|
}
|
|
54
55
|
```
|
|
55
56
|
|
|
@@ -167,7 +168,12 @@ Common `api` values:
|
|
|
167
168
|
|
|
168
169
|
- `thinkingLevelMap` (Pi 0.72) replaces `compat.reasoningEffortMap`. Map pi levels (`off`/`minimal`/`low`/`medium`/`high`/`xhigh`) to provider values; use `null` to hide a level.
|
|
169
170
|
- `openRouterRouting` is forwarded as-is in the OpenRouter `provider` field (fallbacks, ZDR, ignore lists, throughput/latency).
|
|
170
|
-
-
|
|
171
|
+
- `compat.thinkingFormat` supports OpenAI-compatible reasoning variants: `openrouter` sends `reasoning: { effort }`, `together` sends `reasoning: { enabled }` plus `reasoning_effort` when supported, and `qwen-chat-template` targets local Qwen-compatible servers that read `chat_template_kwargs.enable_thinking`.
|
|
172
|
+
- Advanced `compat` flags exist for proxy quirks (`cacheControlFormat`, `supportsReasoningEffort`, `supportsLongCacheRetention`, `supportsEagerToolInputStreaming`, `sendSessionIdHeader`, `sendSessionAffinityHeaders`). See [Pi `docs/models.md`](https://github.com/earendil-works/pi/blob/main/packages/coding-agent/docs/models.md) when a proxy rejects pi's defaults.
|
|
173
|
+
|
|
174
|
+
### Context Overflow Recovery
|
|
175
|
+
|
|
176
|
+
Pi can auto-compact and retry when a provider fails with a recognized context-window overflow. For custom providers whose overflow text Pi does not recognize, use a provider-scoped `message_end` extension handler to rewrite the finalized assistant `errorMessage` so it starts with `context_length_exceeded`.
|
|
171
177
|
|
|
172
178
|
## Custom Providers via Extensions
|
|
173
179
|
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
# Skills
|
|
2
2
|
|
|
3
|
-
Skills are self-contained capability packages that Pi loads on demand. Pi follows the [Agent Skills specification](https://agentskills.io/specification) and stays fairly lenient, warning on
|
|
3
|
+
Skills are self-contained capability packages that Pi loads on demand. Pi follows the [Agent Skills specification](https://agentskills.io/specification) and stays fairly lenient, warning on most violations instead of hard-failing. Pi allows skill names to differ from parent directory names so shared skill directories can work across multiple agent harnesses.
|
|
4
4
|
|
|
5
5
|
## How Skills Work
|
|
6
6
|
|
|
@@ -83,7 +83,7 @@ Read `references/api.md` for the full details.
|
|
|
83
83
|
|
|
84
84
|
| Field | Required | Notes |
|
|
85
85
|
|-------|----------|-------|
|
|
86
|
-
| `name` | Yes | 1-64 chars, lowercase letters/numbers/hyphens
|
|
86
|
+
| `name` | Yes | 1-64 chars, lowercase letters/numbers/hyphens; may differ from parent dir |
|
|
87
87
|
| `description` | Yes | Max 1024 chars; primary trigger text |
|
|
88
88
|
| `license` | No | License name or bundled reference |
|
|
89
89
|
| `compatibility` | No | Environment requirements |
|
|
@@ -113,7 +113,6 @@ To reuse Claude Code or Codex skill directories, add them to settings:
|
|
|
113
113
|
|
|
114
114
|
Pi warns on most issues but still loads the skill when possible:
|
|
115
115
|
|
|
116
|
-
- Name and parent-directory mismatch
|
|
117
116
|
- Name too long or invalid
|
|
118
117
|
- Description too long
|
|
119
118
|
- Duplicate names, where first discovered wins
|