ShellPilot

Author(s)

• raandree

Tags

GitHubCopilot Copilot AI LLM Chat Agent

Functions

Clear-ShpChat Clear-ShpContext Clear-ShpUsage ConvertTo-ShpTokenCount Get-ShpChat Get-ShpContext Get-ShpCosineSimilarity Get-ShpCostEstimate Get-ShpDefault Get-ShpModel Get-ShpModelName Get-ShpTool Get-ShpUsage Initialize-Shp Invoke-Shp Register-ShpTool Request-ShpEmbedding Select-ShpModel Set-ShpContext Start-ShpChat Unregister-ShpTool

PSEditions

Core

Dependencies

This module has no dependencies.

Release Notes

## [0.3.0-preview0003] - 2026-07-09

### Added

- `read_file` now supports bounded, paged reads. The tool schema and
 `Invoke-ReadFileTool` accept optional `offset`/`limit` (a 1-based line window)
 and return a JSON envelope carrying `path`, `totalLines`, `offset`, `limit`,
 `returnedLines`, `hasMore` and the window `text`, so the model can page
 through a large file (read the first window, then request the next while
 `hasMore` is true) instead of loading it whole. Existing `path`-only calls
 keep working and return a bounded first window.
- `Invoke-Shp` now reports `Usage.ContextTokens`: the peak single-request
 prompt size in a turn - how full the model's context window actually got - as
 opposed to `Usage.PromptTokens`, which is the billed sum of input tokens
 across every tool-calling round-trip. For a turn with no tool calls the two
 are equal; for a multi-round-trip turn `ContextTokens` is the largest single
 request's prompt while `PromptTokens` is their sum. The per-call usage record
 carries the field too, and `Get-ShpUsage -Summary` aggregates it as a maximum
 (occupancy does not add across calls) both overall and per model. Purely
 additive - existing token and cost fields (`PromptTokens`, `CompletionTokens`,
 `TotalTokens`, `CachedTokens`, cost) are unchanged.

### Changed

- Every tool result handed back to the model is now bounded by default so one
 large read cannot overflow the context window: `read_file`, `fetch_url` and
 `run_command` output are each capped (default 100,000 characters) with a clear
 `...[truncated, original N chars]` marker. A bare `read_file` returns a bounded
 first window rather than the entire file.
- The default on-disk OAuth token file was renamed from `.copilot-demo-token`
 to `.shellpilot-token` (still a hidden dot-file in the user's home directory).
 The old name dated from ShellPilot's proof-of-concept origin; the new name is
 branded to the module. `-TokenPath` still overrides the location. Because the
 default path changed, existing users must re-run `Initialize-Shp` once to
 write the token under the new name (or pass `-TokenPath` to point at the old
 file); the previous `.copilot-demo-token` file is not migrated automatically
 and can be deleted.

### Fixed

- Reading a large file (or many files) in one turn no longer overflows the model
 context window and fails with `413 Request Entity Too Large` /
 `model_max_prompt_tokens_exceeded`. Alongside the per-result caps above,
 `Invoke-Shp` now guards the context window before each chat request, eliding
 the oldest tool results (via the new private `Compress-ShpChatContext` helper)
 when the estimated prompt exceeds a budget.