agent-postmortem 0.3.4__tar.gz → 0.3.6__tar.gz

{agent_postmortem-0.3.4 → agent_postmortem-0.3.6}/CHANGELOG.md

@@ -5,6 +5,37 @@ All notable changes to this project are documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.3.6]
+
+### Fixed
+
+- The default (share) output could print an agent's own label. A story sentence starts with
+  a capital, and when an agent's whole story was its interrupt finding, whose title starts
+  with the label, the label was capitalized before the name substitution ran. That
+  substitution matched names case-sensitively, so a workflow label such as `impl:t2` or a
+  subagent description starting with a lowercase letter printed as `Impl:t2` in the
+  STORIES section of `run` and in `stories[].text` of `run --json`. A story clause is now
+  capitalized after its names are replaced, so it reads `Agent-1 interrupted after ...`.
+  `scan`, `run --bug-report`, `rescue` and the hook messages never carried it, and none of
+  the 382 runs on the maintainer's machine hit it. `--local` and `--raw` are unchanged.
+
+## [0.3.5]
+
+### Changed
+
+- The PyPI summary is the positioning line the plugin already uses: it reads the Claude Code
+  logs already on your machine and tells you which agent's "done" you can't trust. The package
+  is classified as Alpha, matching the README's status line, and the README's license badge
+  links to the LICENSE file by full URL, so it also works on the PyPI page. PyPI version and
+  Python version badges join the CI and license badges.
+- Installs come from PyPI now. The README's first command is `uvx agent-postmortem scan`, and
+  Quick start, the plugin README and the `/postmortem` skill say `uv tool install
+  agent-postmortem`. No git is needed.
+- The plugin installs its CLI from PyPI at the plugin's own version. It falls back to the
+  matching GitHub tag only if that install fails, for example in the minutes between a release
+  reaching GitHub and reaching PyPI. Both attempts share the one 60-second deadline the single
+  install had. The missing-CLI hint says `uv tool install agent-postmortem`.
+
 ## [0.3.4]
 
 ### Fixed

{agent_postmortem-0.3.4 → agent_postmortem-0.3.6}/CONTRIBUTING.md

@@ -70,22 +70,20 @@ sets.
 
 Releases are cut by pushing a `v*` tag, which `.github/workflows/release.yml`
 picks up: it runs the tests, builds the wheel and sdist, runs the privacy
-check against them, and makes the GitHub release with the built files and
-that version's CHANGELOG section. Users install from that tag
-(`uv tool install git+https://github.com/zeyusuntt/agent-postmortem@v<version>`),
-and so does the plugin.
-
-Publishing to PyPI (via a trusted publisher, no API token stored in this
-repository) is switched off until the repository variable `PUBLISH_PYPI` is
-`true`; without it the publish job is skipped, not failed. To turn it on once
-the setup below is done: set the variable (Settings, Secrets and variables,
-Actions, Variables, `PUBLISH_PYPI` = `true`), then re-run the release workflow
-for the existing tag (Actions, Release, the tag's run, Re-run all jobs), and
-check that installing the `agent-postmortem` package from PyPI gives that
-version. After that,
-switch the install lines in README.md, plugin/README.md, the postmortem skill
-and plugin/hooks/run.sh back to PyPI in one release (tests/test_launch_docs.py
-and tests/test_plugin.py pin them).
+check against them, makes the GitHub release with the built files and
+that version's CHANGELOG section, and publishes both files to PyPI through a
+trusted publisher (no API token is stored in this repository). Users install
+from PyPI (`uv tool install agent-postmortem`), and so does the plugin, pinned
+to its own version, with the matching GitHub tag as a fallback for the minutes
+between the tag and the PyPI upload (tests/test_launch_docs.py and
+tests/test_plugin.py pin these install lines).
+
+The publish job runs only while the repository variable `PUBLISH_PYPI` is
+`true` (Settings, Secrets and variables, Actions, Variables); without it the
+job is skipped, not failed. If a release's publish was skipped, set the
+variable, re-run the release workflow for that tag (Actions, Release, the
+tag's run, Re-run all jobs), and check that installing `agent-postmortem`
+from PyPI gives that version.
 
 One-time trusted-publisher setup on pypi.org, before the first tagged
 release:

{agent_postmortem-0.3.4 → agent_postmortem-0.3.6}/PKG-INFO

@@ -1,13 +1,13 @@
 Metadata-Version: 2.5
 Name: agent-postmortem
-Version: 0.3.4
-Summary: Post-mortem for multi-agent coding runs: prove what silently went wrong, then draft the rule that prevents it.
+Version: 0.3.6
+Summary: Reads the Claude Code logs already on your machine and tells you which agent's 'done' you can't trust, and what to run next.
 Project-URL: Homepage, https://github.com/zeyusuntt/agent-postmortem
 Project-URL: Issues, https://github.com/zeyusuntt/agent-postmortem/issues
 Project-URL: Changelog, https://github.com/zeyusuntt/agent-postmortem/blob/main/CHANGELOG.md
 License: MIT
 License-File: LICENSE
-Classifier: Development Status :: 4 - Beta
+Classifier: Development Status :: 3 - Alpha
 Classifier: Environment :: Console
 Classifier: Intended Audience :: Developers
 Classifier: License :: OSI Approved :: MIT License
@@ -30,15 +30,17 @@ agent-postmortem reads the transcripts Claude Code already wrote under `~/.claud
 On the maintainer's machine (scan 2026-10-27, 0.3.2), 13 of 166 workflow runs launched at least one agent's task a second time, and 10.2 hours went to retries and to attempts that ended without a result.
 
 ```
-uvx --from git+https://github.com/zeyusuntt/agent-postmortem@v0.3.4 agent-postmortem scan
+uvx agent-postmortem scan
 ```
 
-That runs the tagged release straight from GitHub (it needs [uv](https://docs.astral.sh/uv/) and git). A PyPI release is coming. On a machine with no runs yet it says where it looked and suggests `agent-postmortem demo`.
+That runs the latest release from PyPI with [uv](https://docs.astral.sh/uv/), which fetches Python 3.11 or later if you don't have it; nothing else to install. On a machine with no runs yet it says where it looked and suggests `agent-postmortem demo`.
 
 The command-line tool is local and read-only: Python standard library only, no LLM, no network access, and it never changes your files, transcripts, commits, or branches. (The Claude Code plugin's one network action, installing that tool, is described under Privacy; Limits notes two rare git index modes in which git's own `status` still updates its bookkeeping.) What it covers, below, lists the run shapes it handles and the ones it does not yet. Status: alpha, used so far on the maintainer's own machines. Claude Code only.
 
 [![CI](https://github.com/zeyusuntt/agent-postmortem/actions/workflows/ci.yml/badge.svg)](https://github.com/zeyusuntt/agent-postmortem/actions/workflows/ci.yml)
-[![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)
+[![PyPI](https://img.shields.io/pypi/v/agent-postmortem.svg)](https://pypi.org/project/agent-postmortem/)
+[![Python versions](https://img.shields.io/pypi/pyversions/agent-postmortem.svg)](https://pypi.org/project/agent-postmortem/)
+[![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](https://github.com/zeyusuntt/agent-postmortem/blob/main/LICENSE)
 
 ```
 RUN wf_00000000-d30  DEGRADED  ended 2026-01-06 19:55Z
@@ -96,10 +98,10 @@ If you run one of the partial or not-yet shapes, please open a ["my run shape is
 
 ## Quick start
 
-Install the command-line tool from the tagged GitHub release with [uv](https://docs.astral.sh/uv/) and git (uv fetches Python 3.11 or later if you don't have it), then try it:
+Install the command-line tool from PyPI with [uv](https://docs.astral.sh/uv/) (it fetches Python 3.11 or later if you don't have it), then try it:
 
 ```
-uv tool install git+https://github.com/zeyusuntt/agent-postmortem@v0.3.4
+uv tool install agent-postmortem
 agent-postmortem demo
 agent-postmortem scan
 agent-postmortem run <run-id>
@@ -245,7 +247,7 @@ Paths inside free text (finding titles, diagnostics, error messages, and at `--l
 
 State lives in `$XDG_STATE_HOME/agent-postmortem/`, or `~/.local/state/agent-postmortem/` if that variable is unset: `state.json` (which findings a session already reported, and the rules the 0.1 `apply` command wrote, now read-only history), plus `secret` and `aliases.json` for the aliases. All three are mode 0600. The secret and the registry are never printed; `trend` shows the applied-rule history from `state.json` at the active level. If a command prints `stable aliases unavailable` on stderr, that output named everything `agent-1`, `agent-2`, and so on instead of by alias; nothing is repaired automatically, because a new secret would change every alias. Remove both `secret` and `aliases.json` to start over with new aliases.
 
-The plugin's one network action happens on `SessionStart` only: when it has not installed a CLI into its own data directory and no `agent-postmortem` on `PATH` passes its protocol check, it installs the CLI from the GitHub tag that matches the plugin's version (`v0.3.4`) into that data directory, retrying at most once every 24 hours on failure; opt out by setting `AGENT_POSTMORTEM_NO_AUTO_INSTALL=1`, or by installing the CLI yourself first.
+The plugin's one network action happens on `SessionStart` only: when it has not installed a CLI into its own data directory and no `agent-postmortem` on `PATH` passes its protocol check, it installs the CLI from PyPI at the plugin's own version, or from the matching GitHub tag if that install fails, into that data directory, retrying at most once every 24 hours on failure; opt out by setting `AGENT_POSTMORTEM_NO_AUTO_INSTALL=1`, or by installing the CLI yourself first.
 
 ## How it was built
 

{agent_postmortem-0.3.4 → agent_postmortem-0.3.6}/README.md

@@ -7,15 +7,17 @@ agent-postmortem reads the transcripts Claude Code already wrote under `~/.claud
 On the maintainer's machine (scan 2026-10-27, 0.3.2), 13 of 166 workflow runs launched at least one agent's task a second time, and 10.2 hours went to retries and to attempts that ended without a result.
 
 ```
-uvx --from git+https://github.com/zeyusuntt/agent-postmortem@v0.3.4 agent-postmortem scan
+uvx agent-postmortem scan
 ```
 
-That runs the tagged release straight from GitHub (it needs [uv](https://docs.astral.sh/uv/) and git). A PyPI release is coming. On a machine with no runs yet it says where it looked and suggests `agent-postmortem demo`.
+That runs the latest release from PyPI with [uv](https://docs.astral.sh/uv/), which fetches Python 3.11 or later if you don't have it; nothing else to install. On a machine with no runs yet it says where it looked and suggests `agent-postmortem demo`.
 
 The command-line tool is local and read-only: Python standard library only, no LLM, no network access, and it never changes your files, transcripts, commits, or branches. (The Claude Code plugin's one network action, installing that tool, is described under Privacy; Limits notes two rare git index modes in which git's own `status` still updates its bookkeeping.) What it covers, below, lists the run shapes it handles and the ones it does not yet. Status: alpha, used so far on the maintainer's own machines. Claude Code only.
 
 [![CI](https://github.com/zeyusuntt/agent-postmortem/actions/workflows/ci.yml/badge.svg)](https://github.com/zeyusuntt/agent-postmortem/actions/workflows/ci.yml)
-[![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)
+[![PyPI](https://img.shields.io/pypi/v/agent-postmortem.svg)](https://pypi.org/project/agent-postmortem/)
+[![Python versions](https://img.shields.io/pypi/pyversions/agent-postmortem.svg)](https://pypi.org/project/agent-postmortem/)
+[![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](https://github.com/zeyusuntt/agent-postmortem/blob/main/LICENSE)
 
 ```
 RUN wf_00000000-d30  DEGRADED  ended 2026-01-06 19:55Z
@@ -73,10 +75,10 @@ If you run one of the partial or not-yet shapes, please open a ["my run shape is
 
 ## Quick start
 
-Install the command-line tool from the tagged GitHub release with [uv](https://docs.astral.sh/uv/) and git (uv fetches Python 3.11 or later if you don't have it), then try it:
+Install the command-line tool from PyPI with [uv](https://docs.astral.sh/uv/) (it fetches Python 3.11 or later if you don't have it), then try it:
 
 ```
-uv tool install git+https://github.com/zeyusuntt/agent-postmortem@v0.3.4
+uv tool install agent-postmortem
 agent-postmortem demo
 agent-postmortem scan
 agent-postmortem run <run-id>
@@ -222,7 +224,7 @@ Paths inside free text (finding titles, diagnostics, error messages, and at `--l
 
 State lives in `$XDG_STATE_HOME/agent-postmortem/`, or `~/.local/state/agent-postmortem/` if that variable is unset: `state.json` (which findings a session already reported, and the rules the 0.1 `apply` command wrote, now read-only history), plus `secret` and `aliases.json` for the aliases. All three are mode 0600. The secret and the registry are never printed; `trend` shows the applied-rule history from `state.json` at the active level. If a command prints `stable aliases unavailable` on stderr, that output named everything `agent-1`, `agent-2`, and so on instead of by alias; nothing is repaired automatically, because a new secret would change every alias. Remove both `secret` and `aliases.json` to start over with new aliases.
 
-The plugin's one network action happens on `SessionStart` only: when it has not installed a CLI into its own data directory and no `agent-postmortem` on `PATH` passes its protocol check, it installs the CLI from the GitHub tag that matches the plugin's version (`v0.3.4`) into that data directory, retrying at most once every 24 hours on failure; opt out by setting `AGENT_POSTMORTEM_NO_AUTO_INSTALL=1`, or by installing the CLI yourself first.
+The plugin's one network action happens on `SessionStart` only: when it has not installed a CLI into its own data directory and no `agent-postmortem` on `PATH` passes its protocol check, it installs the CLI from PyPI at the plugin's own version, or from the matching GitHub tag if that install fails, into that data directory, retrying at most once every 24 hours on failure; opt out by setting `AGENT_POSTMORTEM_NO_AUTO_INSTALL=1`, or by installing the CLI yourself first.
 
 ## How it was built
 

{agent_postmortem-0.3.4 → agent_postmortem-0.3.6}/docs/DESIGN.md

@@ -4742,7 +4742,19 @@ trend row's target) and `Finding.title`, rendered through `Redactor.title()`
 followed by structural `redact()`; **(N3b S1)** names come first, so a label
 that holds a UUID, an email, or a `-Users-...` slug is replaced whole
 instead of being half-rewritten by the structural pass and then missed)
-work exactly as before. Evidence, a diagnostic, and a metrics value are
+work exactly as before. **(0.3.6)** A story clause is capitalized only after the sink
+substituted names in it (`report._render_pieces` renders each `_Clause` through
+`Redactor.title`, then upper-cases its first character), so a name meets the pattern as
+written and becomes `Agent-1` at a sentence start. Through 0.3.5 the clause was capitalized
+first: a `mislabeled_kill` title starts with the label, and when it was an agent's whole
+story the label `impl:x` reached share output as `Impl:x`, which the case-sensitive pattern
+never matched (`run` text and `run --json` stories only; `tests/test_share_label_case.py`).
+Matching stays case-sensitive, and every clause in the tool's own words is written with
+its capital already (`report._own_clause`: "Killed 1 time by the runtime", "Attempt 2
+failed: ...", "Wrote no paths."), so a label that is a plain word (`killed`, `attempt`,
+`wrote`) never rewrites them, and neither are finding titles, except the title of a
+detector that leads with its agent's own label (`report._LABEL_LED_TITLES`:
+`mislabeled_kill`), which is capitalized only after the substitution. Evidence, a diagnostic, and a metrics value are
 never fed to the five *name* categories (agent/desc/phase/type/target): they
 pass through structural `redact()` alone (plain `Redactor.__call__`),
 because they are a faithful quote of transcript text (4.5), not a rendered
@@ -6377,9 +6389,12 @@ internal deadline as `check`.
 
 On SessionStart, if no compatible CLI resolves and `uv` is on PATH and
 `AGENT_POSTMORTEM_NO_AUTO_INSTALL` is unset, `run.sh` runs `uv tool install
+"agent-postmortem==<version in plugin.json>" --python 3.11+` from PyPI, and only if that
+install fails, `uv tool install
 "git+https://github.com/zeyusuntt/agent-postmortem@v<version in plugin.json>" --python
-3.11+` (Plan B, 2026-10-28: the public repository's tag instead of PyPI until PyPI
-publishing works; it needs git besides uv; the missing-CLI hint names the same source) (`UV_TOOL_DIR`
+3.11+` (0.3.5: PyPI first; the tag covers the minutes between a release reaching GitHub
+and reaching PyPI, and needs git besides uv; both attempts run inside the one 60 s
+deadline below; the missing-CLI hint names `uv tool install agent-postmortem`) (`UV_TOOL_DIR`
 and `UV_TOOL_BIN_DIR` both pointed at `$DATA_DIR`, 7.2's one data
 directory, so the installed shim lands at `$DATA_DIR/bin/agent-postmortem`,
 exactly where resolution step (1) looks) under the same process-group-kill
@@ -6704,7 +6719,7 @@ later events go positional).
 
 ### 8.1 Package
 
-`pyproject.toml`: name `agent-postmortem`, version `0.3.4`, console scripts
+`pyproject.toml`: name `agent-postmortem`, version `0.3.6`, console scripts
 `agent-postmortem` and `postmortem`, `requires-python >= 3.11`, no runtime
 dependencies, `[project.optional-dependencies].dev = ["pytest>=8",
 "ruff>=0.6"]`, classifiers, URLs (Homepage, Issues, Changelog). The wheel
@@ -6754,8 +6769,8 @@ check of the artifacts), then a `github-release` job (`permissions: contents: wr
 `gh release create <tag> dist/* --verify-tag` with that version's CHANGELOG section as
 notes) and a publish job with `environment: pypi`, `permissions: id-token: write`,
 `pypa/gh-action-pypi-publish@release/v1` that runs only when the repository variable
-`PUBLISH_PYPI` is `true` (Plan B, 2026-10-28: until the PyPI trusted publisher works, a
-tag must not turn the release red; a skipped job is not a failure). Both workflows declare a
+`PUBLISH_PYPI` is `true` (a tag must never turn the release red when publishing is
+switched off; a skipped job is not a failure; PyPI publishing is on from 0.3.4). Both workflows declare a
 top-level `permissions: contents: read` so every job is read-only by
 default regardless of the repository's own token-permission setting; the
 publish job's own `id-token: write` is layered on top of that at the job
@@ -6769,14 +6784,14 @@ pypi.org and the TestPyPI dry run.
 First screen, in order (from 0.3.3): name; the question the tool answers
 ("Which 'done' in your last Claude Code run can't you trust?"); one paragraph on
 what it reads, what it reports and that it is local and read-only; one dated number
-from a named version's scan; `uvx --from git+https://github.com/zeyusuntt/agent-postmortem@v<version>
-agent-postmortem scan` (the tagged GitHub release; one line says a PyPI release is coming); the status paragraph; badges (CI, license; PyPI version and Python versions only once a
+from a named version's scan; `uvx agent-postmortem scan` (from PyPI, 0.3.5; 0.3.3 and
+0.3.4 used the tagged GitHub release while PyPI was not yet working); the status paragraph; badges (CI, license; PyPI version and Python versions only once a
 PyPI release exists, since a badge for a missing release renders broken),
 the demo output as a fenced code block (not a screenshot: it must stay text,
 so it can be diffed against a live `agent-postmortem demo` run -- 8.2,
 tests/test_readme.py), the install command that works for the current
-distribution (a GitHub install with `uv tool install git+...` until the
-first PyPI release, then `uvx agent-postmortem scan`), a command table
+distribution (`uv tool install agent-postmortem` from PyPI from 0.3.5; a
+GitHub install with `uv tool install git+...` in 0.3.3 and 0.3.4), a command table
 listing exactly which commands take `--json` and which take `--no-redact`.
 Then "What it covers", before Quick start: one table of run shapes and failure
 kinds, each marked covered, partial or not yet, as checked against the upstream issue

{agent_postmortem-0.3.4 → agent_postmortem-0.3.6}/plugin/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "agent-postmortem",
-  "version": "0.3.4",
+  "version": "0.3.6",
   "description": "Reads the Claude Code logs already on your machine and tells you which agent's 'done' you can't trust, and what to run next.",
   "author": {
     "name": "zeyusuntt",

{agent_postmortem-0.3.4 → agent_postmortem-0.3.6}/plugin/README.md

@@ -41,15 +41,16 @@ On `SessionStart`, if the plugin has not installed a CLI into its own data
 directory and no `agent-postmortem` on your `PATH` passes its protocol check
 (`agent-postmortem hook capabilities` answers protocol 1), and
 [`uv`](https://docs.astral.sh/uv/) is on your `PATH`, the plugin installs
-the version pinned in the plugin manifest from its GitHub tag (`git` is
-needed too), retrying at most once every 24
+the version pinned in the plugin manifest from PyPI, or from the matching
+GitHub tag if that install fails (that fallback needs `git`), retrying at
+most once every 24
 hours on failure, and writes what happened to a log file instead of your
 terminal. This is the plugin's only network access.
 
 Opt out by setting `AGENT_POSTMORTEM_NO_AUTO_INSTALL=1` (in your shell
 profile, or wherever you set environment variables for Claude Code), or by
 installing the CLI yourself ahead of time with
-`uv tool install git+https://github.com/zeyusuntt/agent-postmortem@v0.3.4`.
+`uv tool install agent-postmortem`.
 
 ## The `/postmortem` skill
 

{agent_postmortem-0.3.4 → agent_postmortem-0.3.6}/plugin/hooks/run.sh

@@ -40,7 +40,9 @@ fi
 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" >/dev/null 2>&1 && pwd)"
 PLUGIN_ROOT="$(cd "$SCRIPT_DIR/.." >/dev/null 2>&1 && pwd)"
 PLUGIN_JSON="$PLUGIN_ROOT/.claude-plugin/plugin.json"
-# The CLI install source (docs/DESIGN.md 7.3): the public repository at the tag v<plugin version>.
+# The CLI install sources (docs/DESIGN.md 7.3): PyPI at the plugin version first, then the
+# public repository at the tag v<plugin version> if that install fails.
+PYPI_PACKAGE="agent-postmortem"
 GIT_SOURCE="git+https://github.com/zeyusuntt/agent-postmortem"
 
 # docs/DESIGN.md 7.2's D9: one data directory, derived once, used for
@@ -325,8 +327,8 @@ _print_missing_cli_hint_if_due() {
     if _modified_within_a_day "$marker"; then
         return 0
     fi
-    printf '{"systemMessage": "agent-postmortem CLI not found; run: uv tool install %s@v%s"}\n' \
-        "$GIT_SOURCE" "$(_plugin_version)"
+    printf '{"systemMessage": "agent-postmortem CLI not found; run: uv tool install %s"}\n' \
+        "$PYPI_PACKAGE"
     : >"$marker" 2>/dev/null
 }
 
@@ -353,16 +355,22 @@ _maybe_install() {
     pinned_version="$(_plugin_version)"
     [ -n "$pinned_version" ] || return 0
 
-    # Plan B (2026-10-28): the git tag that matches the plugin version, until PyPI works.
-    source="$GIT_SOURCE@v$pinned_version"
+    # PyPI at the plugin version first; the matching GitHub tag only if that install fails
+    # (for example in the minutes between a release reaching GitHub and reaching PyPI). Both
+    # run under the one 60-second deadline the single install had, so the hook's time budget
+    # is unchanged.
+    pypi_source="$PYPI_PACKAGE==$pinned_version"
+    git_source="$GIT_SOURCE@v$pinned_version"
     {
-        printf '%s attempting: uv tool install %s\n' \
-            "$(date -u +%Y-%m-%dT%H:%M:%SZ)" "$source"
+        printf '%s attempting: uv tool install %s (fallback: %s)\n' \
+            "$(date -u +%Y-%m-%dT%H:%M:%SZ)" "$pypi_source" "$git_source"
     } >>"$install_log" 2>/dev/null
 
     mkdir -p "$DATA_DIR/bin" 2>/dev/null
     UV_TOOL_DIR="$DATA_DIR" UV_TOOL_BIN_DIR="$DATA_DIR/bin" \
-        _run_with_deadline 60 uv tool install "$source" --python 3.11+ \
+        _run_with_deadline 60 sh -c \
+        'uv tool install "$1" --python 3.11+ || uv tool install "$2" --python 3.11+' \
+        sh "$pypi_source" "$git_source" \
         </dev/null >>"$install_log" 2>&1
 }
 

{agent_postmortem-0.3.4 → agent_postmortem-0.3.6}/plugin/skills/postmortem/SKILL.md

@@ -114,5 +114,5 @@ Bash tool.
   `--no-redact` only if the user asks to see raw paths or session ids, and
   say so when you do.
 - If the CLI is missing entirely, tell the user to run
-  `uv tool install git+https://github.com/zeyusuntt/agent-postmortem@v0.3.4` and stop; do not try to install it
+  `uv tool install agent-postmortem` and stop; do not try to install it
   yourself.

{agent_postmortem-0.3.4 → agent_postmortem-0.3.6}/pyproject.toml

@@ -1,13 +1,13 @@
 [project]
 name = "agent-postmortem"
-version = "0.3.4"
-description = "Post-mortem for multi-agent coding runs: prove what silently went wrong, then draft the rule that prevents it."
+version = "0.3.6"
+description = "Reads the Claude Code logs already on your machine and tells you which agent's 'done' you can't trust, and what to run next."
 readme = "README.md"
 license = { text = "MIT" }
 requires-python = ">=3.11"
 dependencies = []
 classifiers = [
-    "Development Status :: 4 - Beta",
+    "Development Status :: 3 - Alpha",
     "Environment :: Console",
     "Intended Audience :: Developers",
     "License :: OSI Approved :: MIT License",

{agent_postmortem-0.3.4 → agent_postmortem-0.3.6}/src/agent_postmortem/__init__.py

@@ -6,4 +6,4 @@ plugin/.claude-plugin/plugin.json's version (docs/DESIGN.md 7.1's D1,
 enforced by tests/test_plugin.py).
 """
 
-__version__ = "0.3.4"
+__version__ = "0.3.6"

{agent_postmortem-0.3.4 → agent_postmortem-0.3.6}/src/agent_postmortem/report.py

@@ -1759,9 +1759,11 @@ def _agent_story_pieces(agent, agent_findings: list[Finding]) -> list:
                 all_runtime = False
         if all_runtime and silences:
             silence_text = ", ".join(silences)
-            lead = f"killed {k} {_plural(k, 'time')} by the runtime ({silence_text} of silence)"
+            lead = _own_clause(
+                f"killed {k} {_plural(k, 'time')} by the runtime ({silence_text} of silence)"
+            )
         else:
-            lead = f"restarted {k} {_plural(k, 'time')}"
+            lead = _own_clause(f"restarted {k} {_plural(k, 'time')}")
 
     parent_stop = False
     if final is not None:
@@ -1773,13 +1775,13 @@ def _agent_story_pieces(agent, agent_findings: list[Finding]) -> list:
     elif lead is not None:
         clauses.append(lead)
     elif parent_stop:
-        clauses.append("stopped by the parent")
+        clauses.append(_own_clause("stopped by the parent"))
 
     for finding in by_detector.get("attempt_failed", []):
         ordinal = _finding_ordinal(finding, attempts_sorted)
         phrase = _cause_phrase(finding.metrics.get("cause", "unattributed"))
         reset_time = finding.metrics.get("reset_time")
-        text = f"attempt {ordinal} failed: {phrase}"
+        text = _own_clause(f"attempt {ordinal} failed: {phrase}")
         if reset_time:
             text += f", resets {reset_time}"
         clauses.append(text)
@@ -1787,7 +1789,9 @@ def _agent_story_pieces(agent, agent_findings: list[Finding]) -> list:
     for finding in by_detector.get("claim_contradiction", []):
         ordinal = _finding_ordinal(finding, attempts_sorted)
         phrases = _claim_contradiction_phrases(finding.metrics)
-        clauses.append(f"attempt {ordinal} completed but reported {'; '.join(phrases)}")
+        clauses.append(
+            _own_clause(f"attempt {ordinal} completed but reported {'; '.join(phrases)}")
+        )
 
     _paths_display, commits_count = _agent_facts_totals(agent)
     path_count, paths_partial = _agent_paths(agent)
@@ -1818,30 +1822,68 @@ def _agent_story_pieces(agent, agent_findings: list[Finding]) -> list:
         clauses.append("; ".join(tooling_bits))
 
     if not clauses and agent_findings:
-        clauses.append(agent_findings[0].title_parts or agent_findings[0].title)
+        clauses.append(_title_clause(agent_findings[0], agent.label))
 
     return _sentences_pieces(clauses)
 
 
+def _sentence_case(text: str) -> str:
+    """`text` with its first character upper-cased: how a story sentence starts."""
+    return text[:1].upper() + text[1:]
+
+
+def _own_clause(text: str) -> str:
+    """A story clause in the tool's own words, capitalized where it is written (0.4 review round
+    2): every clause is capitalized again only after names are substituted, and a label that is
+    the clause's first word (`killed`, `restarted`, `attempt`) must not meet that word in
+    lowercase."""
+    return _sentence_case(text)
+
+
+# Detectors whose title starts with the agent's own label ("<label> interrupted after ...").
+_LABEL_LED_TITLES = frozenset({"mislabeled_kill"})
+
+
+def _title_clause(finding: Finding, label: str | None) -> object:
+    """A finding title as a story clause (0.3.6 review round 3). The title of a detector that
+    leads with the agent's own label stays as written, so the name substitution meets the label
+    before the sentence is capitalized; every other title is in the tool's own words and is
+    capitalized where it is used (`_own_clause`), so a label that is its first word ("the" in
+    "the harness reported this workflow failed") never rewrites it."""
+    if finding.title_parts:
+        return finding.title_parts
+    title = finding.title
+    if (
+        finding.detector in _LABEL_LED_TITLES
+        and isinstance(label, str) and label and title.startswith(label)
+    ):
+        return title
+    return _own_clause(title)
+
+
+class _Clause(str):
+    """A story clause as written, without its final period. Its first character is upper-cased
+    only after a sink rendered it (0.3.6): a clause can start with a user-authored name (a
+    `mislabeled_kill` title starts with the label), and the case-sensitive name substitution
+    must meet that name as written (docs/DESIGN.md 5.1)."""
+
+
 def _sentences(clauses: list[str]) -> str:
     """Join clauses as sentences: each starts with a capital and the whole
     ends with one period. Finding titles start lowercase by convention
     (docs/DESIGN.md 4.3), so this is where a story gets its capitals."""
     parts = [clause.rstrip(".") for clause in clauses if clause]
-    text = ". ".join(c[:1].upper() + c[1:] for c in parts)
+    text = ". ".join(_sentence_case(c) for c in parts)
     return f"{text}." if text else ""
 
 
 def _sentences_pieces(clauses: list) -> list:
-    """`_sentences` as pieces (F fix 7, W7-2): a clause is a string or the `TitleParts` of a
-    finding title that names a path. With no `TitleParts` the result is the one string
-    `_sentences` returns (or no piece for no text). With one, the template is capitalized and
-    stripped of a final period like any clause, and the pieces are the strings around it (merged)
-    and the `TitleParts` themselves, so `_render_pieces` can render the path once and never scan
-    the tool's own words."""
-    if not any(isinstance(clause, TitleParts) for clause in clauses):
-        text = _sentences(clauses)
-        return [text] if text else []
+    """`_sentences` as pieces (F fix 7, W7-2; 0.3.6): one piece per clause, with the ". "
+    between clauses and the final "." as plain strings. A string clause is a `_Clause`, which a
+    sink renders first and capitalizes after (`_render_pieces`), so a name at its start is
+    substituted as written. The `TitleParts` of a finding title that names a path has its
+    template capitalized and stripped of a final period here, since those words are the tool's
+    own, and `_render_pieces` renders the path once and never scans them. No text, no piece."""
     parts: list = []
     for clause in clauses:
         if not clause:
@@ -1849,35 +1891,39 @@ def _sentences_pieces(clauses: list) -> list:
         if isinstance(clause, TitleParts):
             head, placeholder, tail = clause.template.partition("{path}")
             template = head + placeholder + tail.rstrip(".")
-            parts.append(TitleParts(template[:1].upper() + template[1:], clause.path))
+            parts.append(TitleParts(_sentence_case(template), clause.path))
         else:
-            text = clause.rstrip(".")
-            parts.append(text[:1].upper() + text[1:])
+            parts.append(_Clause(clause.rstrip(".")))
+    if all(isinstance(part, _Clause) and not part for part in parts):
+        return []
     pieces: list = []
     for index, part in enumerate(parts):
-        for piece in ((". ",) if index else ()) + (part,):
-            if isinstance(piece, str) and pieces and isinstance(pieces[-1], str):
-                pieces[-1] += piece
-            else:
-                pieces.append(piece)
-    if pieces and isinstance(pieces[-1], str):
-        pieces[-1] += "."
-    else:
-        pieces.append(".")
+        if index:
+            pieces.append(". ")
+        pieces.append(part)
+    pieces.append(".")
     return pieces
 
 
 def _pieces_raw(pieces: list) -> str:
-    """The pieces with the real path (the raw text of a story)."""
-    return "".join(piece.raw() if isinstance(piece, TitleParts) else piece for piece in pieces)
+    """The pieces with the real path and real names (the raw text of a story)."""
+    return "".join(
+        piece.raw() if isinstance(piece, TitleParts)
+        else _sentence_case(piece) if isinstance(piece, _Clause)
+        else piece
+        for piece in pieces
+    )
 
 
 def _render_pieces(pieces: list, redactor: Redactor) -> str:
-    """The pieces as a sink prints them: a string through `Redactor.title` (labels, paths and the
-    rest of the scanner), a `TitleParts` through `Redactor.title_parts` (its path rendered once,
-    its words never scanned)."""
+    """The pieces as a sink prints them: a `_Clause` through `Redactor.title` (labels, paths and
+    the rest of the scanner) and only then capitalized, a `TitleParts` through
+    `Redactor.title_parts` (its path rendered once, its words never scanned), and the separators
+    as they are."""
     return "".join(
-        redactor.title_parts(piece) if isinstance(piece, TitleParts) else redactor.title(piece)
+        redactor.title_parts(piece) if isinstance(piece, TitleParts)
+        else _sentence_case(redactor.title(piece)) if isinstance(piece, _Clause)
+        else piece
         for piece in pieces
     )
 
@@ -1933,7 +1979,7 @@ def _story_records(run: Run, findings: list[Finding], verbose: bool) -> list[dic
     run_relevant = [f for f in run_level if verbose or f.severity != "info"]
     if run_relevant:
         rank = min(_SEVERITIES.index(f.severity) for f in run_relevant)
-        pieces = _sentences_pieces([f.title_parts or f.title for f in run_relevant])
+        pieces = _sentences_pieces([_title_clause(f, None) for f in run_relevant])
         entries.append({
             "agent": "run",
             "severity": _SEVERITIES[rank],

{agent_postmortem-0.3.4 → agent_postmortem-0.3.6}/tests/synthetic-ids.txt

@@ -545,3 +545,4 @@ wf_00000000-8c0
 wf_00000000-cab
 wf_00000000-cac
 wf_00000000-cad
+wf_00000000-b36