From 5151063c4f49abd93b1056086eb3b3715bad9c3e Mon Sep 17 00:00:00 2001 From: Claude Date: Mon, 7 Sep 2026 15:40:36 +0000 Subject: [PATCH] docs: point the quickstart and the announcement at the PyPI release 0.1.0 is on PyPI, so the README quickstart and the announcement copy no longer need the "from a checkout" install. The announcement also names the two things that landed since it was drafted: the library entry points a catalog calls directly, and a run manifest reaching any OpenAI-compatible endpoint with every run kept in the history. Co-Authored-By: Claude Fable 5.1 Claude-Session: https://claude.ai/code/session_014gwmBVUHSgohCLkNVqAR77 Signed-off-by: Claude --- README.md | 2 +- docs/announcement.md | 9 +++++++-- 2 files changed, 8 insertions(+), 3 deletions(-) diff --git a/README.md b/README.md index 619c678..2ccb86d 100644 --- a/README.md +++ b/README.md @@ -22,7 +22,7 @@ whole (the consumer sets its own thresholds). ## 30-second quickstart ```bash -pip install context-report # once published; today: pip install -e ".[dev]" from a checkout +pip install context-report # 0.1.0 on PyPI; pip install -e ".[dev]" from a checkout to hack on it context-report produce --subject ./my-plugin --kind plugin --target claude_code --n 20 \ --out report.json # one statement: reachability, cost and fault rows for one target context-report run run.json # a whole manifest: subjects x models x tasks in one shot diff --git a/docs/announcement.md b/docs/announcement.md index 19f7b0a..7cc9e1d 100644 --- a/docs/announcement.md +++ b/docs/announcement.md @@ -41,10 +41,14 @@ proof an artifact works everywhere. The fault rows for what a real client does o timeout are `NotAvailable` in v0.1 — it does not yet drive a live client, and says so rather than guessing from documentation. -**Try it in five minutes**: `pip install -e ".[dev]"` from a checkout, then +**Try it in five minutes**: `pip install context-report` (0.1.0 is on PyPI), then `context-report produce --subject ./your-artifact --kind plugin --target claude_code --n 20 --out report.json` produces a statement against your own plugin, hook, or instruction file; `context-report -verify report.json --subject ./your-artifact` checks it against the schema. +verify report.json --subject ./your-artifact` checks it against the schema. The same functions are +a Python library (`from context_report import produce_statement, verify`) for a catalog or CI job. +A run manifest lines up any model you can reach: Anthropic, the `claude` CLI, or any +OpenAI-compatible endpoint, hosted or local, with a `baseUrl`; every run is kept and the history +renders side by side so you can see what a change to the artifact did. **How to help**: measure your own artifact and open the resulting statement as a PR to `spec/attestation/v0.1/examples/`; add a target agent's payload shape to @@ -77,4 +81,5 @@ Every row is `re-derivable` (a catalog can recompute it) or `claimed` (labeled, no thresholds, no certification, just facts. Measured 88 bundles, 18 public plugins, and 7 instruction files so far — three plugins turned out unreachable (exit 126, no execute bit) and every hook allowed malformed input. +`pip install context-report` to try it on your own artifact. Repo: https://github.com/open-coder-ai/context-report — spec feedback and your own reports welcome.