foretop-charter 0.1.0__tar.gz

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.
Files changed (44) hide show
  1. foretop_charter-0.1.0/LICENSE +202 -0
  2. foretop_charter-0.1.0/PKG-INFO +355 -0
  3. foretop_charter-0.1.0/README.md +340 -0
  4. foretop_charter-0.1.0/pyproject.toml +73 -0
  5. foretop_charter-0.1.0/pyproject.toml.orig +89 -0
  6. foretop_charter-0.1.0/src/charter/__init__.py +1 -0
  7. foretop_charter-0.1.0/src/charter/__main__.py +4 -0
  8. foretop_charter-0.1.0/src/charter/brand.py +6 -0
  9. foretop_charter-0.1.0/src/charter/canonical.py +35 -0
  10. foretop_charter-0.1.0/src/charter/capability.py +144 -0
  11. foretop_charter-0.1.0/src/charter/cli.py +232 -0
  12. foretop_charter-0.1.0/src/charter/collect.py +33 -0
  13. foretop_charter-0.1.0/src/charter/drift.py +119 -0
  14. foretop_charter-0.1.0/src/charter/enumerate.py +266 -0
  15. foretop_charter-0.1.0/src/charter/findings.py +293 -0
  16. foretop_charter-0.1.0/src/charter/git_ref.py +30 -0
  17. foretop_charter-0.1.0/src/charter/lockfile.py +109 -0
  18. foretop_charter-0.1.0/src/charter/models.py +60 -0
  19. foretop_charter-0.1.0/src/charter/parsers/__init__.py +0 -0
  20. foretop_charter-0.1.0/src/charter/parsers/claude_code.py +100 -0
  21. foretop_charter-0.1.0/src/charter/parsers/cursor.py +88 -0
  22. foretop_charter-0.1.0/src/charter/render/__init__.py +0 -0
  23. foretop_charter-0.1.0/src/charter/render/annotations.py +53 -0
  24. foretop_charter-0.1.0/src/charter/render/json_renderer.py +14 -0
  25. foretop_charter-0.1.0/src/charter/render/markdown.py +100 -0
  26. foretop_charter-0.1.0/src/charter/render/sarif.py +92 -0
  27. foretop_charter-0.1.0/src/charter/render/terminal.py +131 -0
  28. foretop_charter-0.1.0/src/charter/rules/capability_taxonomy.yaml +216 -0
  29. foretop_charter-0.1.0/src/charter/triage.py +140 -0
  30. foretop_charter-0.1.0/src/keel/__init__.py +1 -0
  31. foretop_charter-0.1.0/src/keel/collect/__init__.py +0 -0
  32. foretop_charter-0.1.0/src/keel/collect/line_tracking.py +48 -0
  33. foretop_charter-0.1.0/src/keel/finding.py +105 -0
  34. foretop_charter-0.1.0/src/keel/gate.py +152 -0
  35. foretop_charter-0.1.0/src/keel/git_ref.py +45 -0
  36. foretop_charter-0.1.0/src/keel/owner.py +58 -0
  37. foretop_charter-0.1.0/src/keel/render/__init__.py +0 -0
  38. foretop_charter-0.1.0/src/keel/render/annotations.py +62 -0
  39. foretop_charter-0.1.0/src/keel/render/markdown.py +40 -0
  40. foretop_charter-0.1.0/src/keel/render/terminal.py +38 -0
  41. foretop_charter-0.1.0/src/keel/render/triage_json.py +15 -0
  42. foretop_charter-0.1.0/src/keel/report.py +98 -0
  43. foretop_charter-0.1.0/src/keel/run.py +67 -0
  44. foretop_charter-0.1.0/src/keel/triage.py +167 -0
@@ -0,0 +1,202 @@
1
+
2
+ Apache License
3
+ Version 2.0, January 2004
4
+ http://www.apache.org/licenses/
5
+
6
+ TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
7
+
8
+ 1. Definitions.
9
+
10
+ "License" shall mean the terms and conditions for use, reproduction,
11
+ and distribution as defined by Sections 1 through 9 of this document.
12
+
13
+ "Licensor" shall mean the copyright owner or entity authorized by
14
+ the copyright owner that is granting the License.
15
+
16
+ "Legal Entity" shall mean the union of the acting entity and all
17
+ other entities that control, are controlled by, or are under common
18
+ control with that entity. For the purposes of this definition,
19
+ "control" means (i) the power, direct or indirect, to cause the
20
+ direction or management of such entity, whether by contract or
21
+ otherwise, or (ii) ownership of fifty percent (50%) or more of the
22
+ outstanding shares, or (iii) beneficial ownership of such entity.
23
+
24
+ "You" (or "Your") shall mean an individual or Legal Entity
25
+ exercising permissions granted by this License.
26
+
27
+ "Source" form shall mean the preferred form for making modifications,
28
+ including but not limited to software source code, documentation
29
+ source, and configuration files.
30
+
31
+ "Object" form shall mean any form resulting from mechanical
32
+ transformation or translation of a Source form, including but
33
+ not limited to compiled object code, generated documentation,
34
+ and conversions to other media types.
35
+
36
+ "Work" shall mean the work of authorship, whether in Source or
37
+ Object form, made available under the License, as indicated by a
38
+ copyright notice that is included in or attached to the work
39
+ (an example is provided in the Appendix below).
40
+
41
+ "Derivative Works" shall mean any work, whether in Source or Object
42
+ form, that is based on (or derived from) the Work and for which the
43
+ editorial revisions, annotations, elaborations, or other modifications
44
+ represent, as a whole, an original work of authorship. For the purposes
45
+ of this License, Derivative Works shall not include works that remain
46
+ separable from, or merely link (or bind by name) to the interfaces of,
47
+ the Work and Derivative Works thereof.
48
+
49
+ "Contribution" shall mean any work of authorship, including
50
+ the original version of the Work and any modifications or additions
51
+ to that Work or Derivative Works thereof, that is intentionally
52
+ submitted to Licensor for inclusion in the Work by the copyright owner
53
+ or by an individual or Legal Entity authorized to submit on behalf of
54
+ the copyright owner. For the purposes of this definition, "submitted"
55
+ means any form of electronic, verbal, or written communication sent
56
+ to the Licensor or its representatives, including but not limited to
57
+ communication on electronic mailing lists, source code control systems,
58
+ and issue tracking systems that are managed by, or on behalf of, the
59
+ Licensor for the purpose of discussing and improving the Work, but
60
+ excluding communication that is conspicuously marked or otherwise
61
+ designated in writing by the copyright owner as "Not a Contribution."
62
+
63
+ "Contributor" shall mean Licensor and any individual or Legal Entity
64
+ on behalf of whom a Contribution has been received by Licensor and
65
+ subsequently incorporated within the Work.
66
+
67
+ 2. Grant of Copyright License. Subject to the terms and conditions of
68
+ this License, each Contributor hereby grants to You a perpetual,
69
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
70
+ copyright license to reproduce, prepare Derivative Works of,
71
+ publicly display, publicly perform, sublicense, and distribute the
72
+ Work and such Derivative Works in Source or Object form.
73
+
74
+ 3. Grant of Patent License. Subject to the terms and conditions of
75
+ this License, each Contributor hereby grants to You a perpetual,
76
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
77
+ (except as stated in this section) patent license to make, have made,
78
+ use, offer to sell, sell, import, and otherwise transfer the Work,
79
+ where such license applies only to those patent claims licensable
80
+ by such Contributor that are necessarily infringed by their
81
+ Contribution(s) alone or by combination of their Contribution(s)
82
+ with the Work to which such Contribution(s) was submitted. If You
83
+ institute patent litigation against any entity (including a
84
+ cross-claim or counterclaim in a lawsuit) alleging that the Work
85
+ or a Contribution incorporated within the Work constitutes direct
86
+ or contributory patent infringement, then any patent licenses
87
+ granted to You under this License for that Work shall terminate
88
+ as of the date such litigation is filed.
89
+
90
+ 4. Redistribution. You may reproduce and distribute copies of the
91
+ Work or Derivative Works thereof in any medium, with or without
92
+ modifications, and in Source or Object form, provided that You
93
+ meet the following conditions:
94
+
95
+ (a) You must give any other recipients of the Work or
96
+ Derivative Works a copy of this License; and
97
+
98
+ (b) You must cause any modified files to carry prominent notices
99
+ stating that You changed the files; and
100
+
101
+ (c) You must retain, in the Source form of any Derivative Works
102
+ that You distribute, all copyright, patent, trademark, and
103
+ attribution notices from the Source form of the Work,
104
+ excluding those notices that do not pertain to any part of
105
+ the Derivative Works; and
106
+
107
+ (d) If the Work includes a "NOTICE" text file as part of its
108
+ distribution, then any Derivative Works that You distribute must
109
+ include a readable copy of the attribution notices contained
110
+ within such NOTICE file, excluding those notices that do not
111
+ pertain to any part of the Derivative Works, in at least one
112
+ of the following places: within a NOTICE text file distributed
113
+ as part of the Derivative Works; within the Source form or
114
+ documentation, if provided along with the Derivative Works; or,
115
+ within a display generated by the Derivative Works, if and
116
+ wherever such third-party notices normally appear. The contents
117
+ of the NOTICE file are for informational purposes only and
118
+ do not modify the License. You may add Your own attribution
119
+ notices within Derivative Works that You distribute, alongside
120
+ or as an addendum to the NOTICE text from the Work, provided
121
+ that such additional attribution notices cannot be construed
122
+ as modifying the License.
123
+
124
+ You may add Your own copyright statement to Your modifications and
125
+ may provide additional or different license terms and conditions
126
+ for use, reproduction, or distribution of Your modifications, or
127
+ for any such Derivative Works as a whole, provided Your use,
128
+ reproduction, and distribution of the Work otherwise complies with
129
+ the conditions stated in this License.
130
+
131
+ 5. Submission of Contributions. Unless You explicitly state otherwise,
132
+ any Contribution intentionally submitted for inclusion in the Work
133
+ by You to the Licensor shall be under the terms and conditions of
134
+ this License, without any additional terms or conditions.
135
+ Notwithstanding the above, nothing herein shall supersede or modify
136
+ the terms of any separate license agreement you may have executed
137
+ with Licensor regarding such Contributions.
138
+
139
+ 6. Trademarks. This License does not grant permission to use the trade
140
+ names, trademarks, service marks, or product names of the Licensor,
141
+ except as required for reasonable and customary use in describing the
142
+ origin of the Work and reproducing the content of the NOTICE file.
143
+
144
+ 7. Disclaimer of Warranty. Unless required by applicable law or
145
+ agreed to in writing, Licensor provides the Work (and each
146
+ Contributor provides its Contributions) on an "AS IS" BASIS,
147
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
148
+ implied, including, without limitation, any warranties or conditions
149
+ of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
150
+ PARTICULAR PURPOSE. You are solely responsible for determining the
151
+ appropriateness of using or redistributing the Work and assume any
152
+ risks associated with Your exercise of permissions under this License.
153
+
154
+ 8. Limitation of Liability. In no event and under no legal theory,
155
+ whether in tort (including negligence), contract, or otherwise,
156
+ unless required by applicable law (such as deliberate and grossly
157
+ negligent acts) or agreed to in writing, shall any Contributor be
158
+ liable to You for damages, including any direct, indirect, special,
159
+ incidental, or consequential damages of any character arising as a
160
+ result of this License or out of the use or inability to use the
161
+ Work (including but not limited to damages for loss of goodwill,
162
+ work stoppage, computer failure or malfunction, or any and all
163
+ other commercial damages or losses), even if such Contributor
164
+ has been advised of the possibility of such damages.
165
+
166
+ 9. Accepting Warranty or Additional Liability. While redistributing
167
+ the Work or Derivative Works thereof, You may choose to offer,
168
+ and charge a fee for, acceptance of support, warranty, indemnity,
169
+ or other liability obligations and/or rights consistent with this
170
+ License. However, in accepting such obligations, You may act only
171
+ on Your own behalf and on Your sole responsibility, not on behalf
172
+ of any other Contributor, and only if You agree to indemnify,
173
+ defend, and hold each Contributor harmless for any liability
174
+ incurred by, or claims asserted against, such Contributor by reason
175
+ of your accepting any such warranty or additional liability.
176
+
177
+ END OF TERMS AND CONDITIONS
178
+
179
+ APPENDIX: How to apply the Apache License to your work.
180
+
181
+ To apply the Apache License to your work, attach the following
182
+ boilerplate notice, with the fields enclosed by brackets "[]"
183
+ replaced with your own identifying information. (Don't include
184
+ the brackets!) The text should be enclosed in the appropriate
185
+ comment syntax for the file format. We also recommend that a
186
+ file or class name and description of purpose be included on the
187
+ same "printed page" as the copyright notice for easier
188
+ identification within third-party archives.
189
+
190
+ Copyright 2026 the foretop authors
191
+
192
+ Licensed under the Apache License, Version 2.0 (the "License");
193
+ you may not use this file except in compliance with the License.
194
+ You may obtain a copy of the License at
195
+
196
+ http://www.apache.org/licenses/LICENSE-2.0
197
+
198
+ Unless required by applicable law or agreed to in writing, software
199
+ distributed under the License is distributed on an "AS IS" BASIS,
200
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
201
+ See the License for the specific language governing permissions and
202
+ limitations under the License.
@@ -0,0 +1,355 @@
1
+ Metadata-Version: 2.4
2
+ Name: foretop-charter
3
+ Version: 0.1.0
4
+ Summary: Scans an agent configuration and produces a reviewable record of every MCP server, tool, scope and credential path — then fails CI when an update quietly expands what the agent can reach.
5
+ License-Expression: Apache-2.0
6
+ License-File: LICENSE
7
+ Requires-Dist: typer>=0.15
8
+ Requires-Dist: rich>=13.9
9
+ Requires-Dist: pyyaml>=6.0
10
+ Requires-Dist: pydantic>=2.9
11
+ Requires-Dist: pathspec>=0.12
12
+ Requires-Dist: httpx>=0.28
13
+ Requires-Python: >=3.12
14
+ Description-Content-Type: text/markdown
15
+
16
+ # charter
17
+
18
+ > Scans an agent configuration and produces a reviewable record of every MCP server, tool,
19
+ > scope and credential path — then fails CI when an update quietly expands what the agent can
20
+ > reach.
21
+
22
+ Full product spec: `../../specs/charter.md`. This file tracks build status; the spec is the
23
+ source of truth for scope and decisions.
24
+
25
+ **Scope note (Session 15):** specs/charter.md's own §0 describes a much larger build —
26
+ Supabase-backed org inventory, GitHub OAuth, Paddle/Lemon Squeezy billing, a hosted dashboard
27
+ (FR-007, its own Slice 5). None of that is in scope here, same boundary ebb and telltale both
28
+ shipped within: the OSS CLI + CI Action is the whole product until `PILOT_STATUS >= 3`
29
+ (specs/charter.md's own "Validation gate"). Sessions 15-17 build exactly the pattern
30
+ CLAUDE_CODE_PLAN.md's Part 4 describes: collector, distinctive engine, renderers + golden
31
+ corpus, Action + PR comment + publish-prep.
32
+
33
+ **Decisions locked in from specs/charter.md §1's checkpoint** (all defaults except DEC-03,
34
+ which the user picked explicitly):
35
+ - **DEC-01** (config sources): Claude Code's `.mcp.json` and Cursor's `.cursor/mcp.json`, both
36
+ project-root, both meant to be committed — matches DEC-04's "the review is a code review"
37
+ framing. Claude Desktop's `claude_desktop_config.json` lives in a per-machine app-support
38
+ directory, never committed to a repo, so it's out of scope for a "review this in CI" tool.
39
+ - **DEC-02** (tool inventories): static parse by default; live `tools/list` enumeration behind
40
+ an explicit `--enumerate` flag, since that means launching third-party code.
41
+ - **DEC-03** (classification): rules-only for v1 — no LLM dependency, no API key, no cost. The
42
+ spec's own default is hybrid (rules + a cached LLM assist for unrecognized tools); that part
43
+ is deliberately not built yet, same as every other real-external-dependency decision this
44
+ build has paused on rather than deciding unilaterally. An unrecognized tool is classified
45
+ `unknown` — a real, honest verdict — not guessed at.
46
+ - **DEC-04** (diff baseline): a committed `charter.lock` file.
47
+ - **DEC-05** (policy language): YAML.
48
+ - **DEC-06** (secret detection depth): names and paths only. This is a hard invariant, not
49
+ just a convention — see below.
50
+
51
+ ## Status
52
+
53
+ **Slice 1 (Session 15): config parse + manifest write + the stable-hash property test.**
54
+
55
+ ```
56
+ uv run --package foretop-charter charter scan <path>
57
+ ```
58
+
59
+ Checks a small, fixed set of committed, project-scoped config paths under `path` —
60
+ `.mcp.json` (Claude Code) and `.cursor/mcp.json` (Cursor), both confirmed live against each
61
+ client's own docs this session, not a full repo walk (`src/charter/collect.py`; same reasoning
62
+ telltale's own OpenAPI-spec autodetection uses for a fixed candidate list). Parses whichever
63
+ exist into `Server` records (`src/charter/models.py`) and writes a canonical, deterministic
64
+ `charter.lock` (`src/charter/lockfile.py`) — the committed diff baseline DEC-04 calls for:
65
+ "it makes the review a code review, which is the whole product." No policy engine or differ
66
+ yet (that's Slice 3) — for now, git's own diff on the committed lock file *is* the review
67
+ mechanism, the same way a `package-lock.json` works.
68
+
69
+ **Canonical serialization** (`src/charter/canonical.py`) — specs/charter.md §6: "sorted keys,
70
+ no floats, LF endings, trailing newline. Test byte-equality across two runs." Recursive
71
+ `sort_keys=True` makes dict-insertion-order irrelevant; a value that would serialize unstably
72
+ (a float) is rejected outright rather than trusted to `repr()` consistently; the lock file is
73
+ written with `Path.write_bytes`, never a text-mode handle, so Windows can't silently turn `\n`
74
+ into `\r\n`. Proven with a genuine Hypothesis property test
75
+ (`tests/unit/test_stable_hash_property.py`), not just a fixed fixture run twice: generates
76
+ ~100 arbitrary (valid-shaped) `.mcp.json` documents per run and asserts the rendered manifest
77
+ is byte-identical across two scans of the same unchanged file.
78
+
79
+ **DEC-06 as a hard invariant, not a convention**: `Server` has no field that could carry a
80
+ credential's *value* — `env_var_names`/`header_names` are exactly that, names only (evidence
81
+ that a server references a credential, e.g. `AIRTABLE_API_KEY`), never what's assigned to
82
+ them. This is structural (there's nowhere on the dataclass to put a value even by accident),
83
+ not just tested — though it's tested too, at the parser, lockfile, and CLI layers, each
84
+ asserting a planted fake secret value never appears in output. `args` is the one field
85
+ captured verbatim rather than name-only, a deliberate choice made after a real fixture
86
+ surfaced why it matters — see `Server`'s own docstring in `models.py` for the full reasoning
87
+ (short version: a positional CLI argument can itself be a connection string with a literal
88
+ password in it, and DEC-06's own default is "never attempt to detect actual secret values" —
89
+ which applies to `args` exactly as much as it applies to `env`/`headers`; hiding `args` would
90
+ also hide the launch command a reviewer needs to actually understand what a server does).
91
+
92
+ **Two client config formats, confirmed live, not from memory**: Claude Code's `.mcp.json`
93
+ (`code.claude.com/docs/en/mcp`) requires an explicit `type` on any `url`-having entry — Claude
94
+ Code itself treats a bare `url` with no `type` as a configuration error and skips that server;
95
+ `src/charter/parsers/claude_code.py` mirrors that exact real behavior rather than guessing a
96
+ transport Claude Code itself wouldn't have used. Cursor's `.cursor/mcp.json`
97
+ (`cursor.com/docs/context/mcp`) has no `type` field at all — presence of `command` vs. `url`
98
+ is what distinguishes stdio from remote, and Cursor's own docs don't expose a separate SSE/
99
+ WebSocket distinction the way Claude Code's `type` does, so every Cursor remote entry is
100
+ recorded as `http`, a documented limitation of what the format itself reveals. `envFile` and
101
+ `auth` (real fields in Cursor's format) aren't read yet — left for a later slice rather than
102
+ half-modeled now.
103
+
104
+ **`keel.collect.line_tracking`**: charter's two parsers get real `evidence_line` values from
105
+ the exact same line-tracking YAML/JSON loader telltale's own parsers already used — extracted
106
+ from `apps/telltale/src/telltale/yaml_lines.py` into `packages/keel` this session, since
107
+ charter's need for it was genuinely identical, not merely similar (see
108
+ `packages/keel/README.md`'s own Session 15 note for the full extraction story).
109
+
110
+ **Verified for real** against a fixture covering all three interesting cases at once — a
111
+ stdio server with a real credential env var, a remote HTTP server with a real credential
112
+ header, and a `url`-with-no-`type` misconfigured entry — before any of this shipped: ran
113
+ `charter scan` against it, confirmed the misconfigured entry was correctly skipped, confirmed
114
+ the terminal output and `charter.lock` both showed `AIRTABLE_API_KEY`/`Authorization` as
115
+ *names* and never the planted fake secret *values*, and confirmed a Cursor-format server in
116
+ `.cursor/mcp.json` with a connection-string-in-args was captured (which is what surfaced the
117
+ `args`-verbatim design note above).
118
+
119
+ **Slice 2, part 1 (Session 15): live enumeration.**
120
+
121
+ ```
122
+ uv run --package foretop-charter charter scan <path> --enumerate
123
+ ```
124
+
125
+ DEC-02: "launching third-party servers is a security decision the user must make consciously"
126
+ — static parsing (Slice 1) never launches anything; `--enumerate` does, so it's opt-in and
127
+ never the default. Stdio servers only for now — remote (http/sse/ws) enumeration would mean
128
+ building an OAuth-capable HTTP+SSE MCP client, real scope creep beyond what this slice needs;
129
+ a documented gap, not a silent one.
130
+
131
+ `src/charter/enumerate.py` is a real MCP client speaking the classic
132
+ `initialize` → `notifications/initialized` → `tools/list` handshake over newline-delimited
133
+ JSON-RPC on the child's stdin/stdout — built from `modelcontextprotocol.io`'s own lifecycle
134
+ and stdio-transport spec pages, fetched live this session, not from memory. Deliberately pinned
135
+ to protocol version `2025-06-18` rather than the `2026-07-28` draft's newer
136
+ `server/discover`-based "modern" era: essentially no real-world server speaks the new one yet,
137
+ and specs/charter.md §11 already names protocol churn as an ongoing risk to revisit, not solve
138
+ preemptively.
139
+
140
+ **A hard timeout, and a real one** — a background thread runs the whole exchange, an overall
141
+ `queue.Queue.get(timeout=...)` bounds it, and shutdown always follows the spec's own sequence
142
+ (close stdin → wait → `SIGTERM` → `SIGKILL`). Verified for real, not asserted: a dedicated
143
+ `--sleep-forever` mode in the test fixture below proves a hung server is actually killed within
144
+ the timeout, not left running — `ps aux` was checked by hand after the first run to confirm no
145
+ orphaned process was left behind before this shipped.
146
+
147
+ **`${VAR}`/`${VAR:-default}` expansion** — Claude Code's own documented syntax
148
+ (`code.claude.com/docs/en/mcp`, fetched live). A real server needs its real `command`/`args`/
149
+ `env` values to actually launch, which `Server` itself never carries (DEC-06). `_resolve_launch`
150
+ re-reads the original config entry for exactly the one server being enumerated and resolves
151
+ values only as local variables inside `enumerate.py`'s own call stack, passed straight into
152
+ `subprocess.Popen`'s `env=` — never assigned to any field, logged, or serialized anywhere else.
153
+ Confirmed by a dedicated test that plants a fake secret value and asserts it never appears in
154
+ `repr()` of the result.
155
+
156
+ **`tests/fixtures/toy_mcp_server.py`**: a real, minimal MCP server, hand-written from the same
157
+ spec pages `enumerate.py` was built against — no MCP SDK, no charter code — so a passing test
158
+ against it is evidence the wire protocol handling actually works against a real subprocess
159
+ exchanging real messages, not just that two halves of a mock agree with each other. Supports
160
+ `--fail-tools-list`, `--crash-on-init`, `--sleep-forever` flags to exercise every failure mode
161
+ `enumerate_stdio_server` has to degrade through.
162
+
163
+ **Lock file schema v2**: adds `tools`/`enumeration_error` per server, three states mirroring
164
+ telltale's own three-state coverage discipline — `tools: null` means not attempted (a plain
165
+ scan, or this server isn't stdio), `tools: []` means attempted and the server genuinely
166
+ advertised none, and a real tools array means it succeeded. Never conflating "we didn't check"
167
+ with "we checked and found nothing" is the same principle either way.
168
+
169
+ Verified end-to-end for real: `charter scan --enumerate` against a config pointing at the toy
170
+ server, confirming the terminal output and `charter.lock` both show the two real discovered
171
+ tools (`read_file`, `write_file`) with their real input schemas.
172
+
173
+ **Slice 2, part 2 (Session 15): the rules-only capability classifier.**
174
+
175
+ `src/charter/capability.py` classifies every tool live enumeration discovers into
176
+ specs/charter.md §2.2's exact five categories (`read`/`write`/`network_egress`/
177
+ `code_execution`/`credential_access`) — DEC-03's rules-only choice, no LLM. A tool can match
178
+ more than one: `web_search` is both `read` (returns information) and `network_egress` (makes
179
+ an outbound call) — under-reporting either would hide something real. Overall severity is the
180
+ *maximum* across every capability a tool matched (`credential_access`/`code_execution` →
181
+ CRITICAL, `write` → HIGH, `network_egress` → MEDIUM, `read` → LOW), per specs/charter.md's own
182
+ accuracy strategy: "errs toward higher severity; a misclassified-as-dangerous tool is
183
+ annoying, the reverse is a breach." An unrecognized tool matches nothing and gets `unknown`
184
+ (`capabilities: []`, severity MEDIUM) — a real, honest verdict, not guessed at, and not
185
+ silently treated as safe either.
186
+
187
+ Rules live in one versioned YAML file with a changelog
188
+ (`src/charter/rules/capability_taxonomy.yaml`, specs/charter.md §6's explicit requirement),
189
+ matched by substring on the tool's lowercased name/description and by lowercased input-schema
190
+ property-name membership — deliberately simple, the same "good enough, not semantically
191
+ perfect" scope call telltale's own PromQL/OTTL condition matching already makes. Patterns were
192
+ derived from real tool-naming conventions seen across the official filesystem/GitHub/Slack/
193
+ Postgres MCP servers and the `tools/list` spec's own examples, not invented from nothing. A
194
+ real accuracy floor (golden corpus, precision/recall) is Slice 3's job, not this file's.
195
+
196
+ Lock file schema v3 adds `capabilities`/`severity`/`rule_version` per tool — capabilities are
197
+ sorted before serializing, not left in `frozenset` iteration order, which Python does not
198
+ guarantee stable across separate process runs (string-hash randomization);
199
+ `test_capabilities_are_sorted_for_byte_stability` guards this directly.
200
+
201
+ **A real bug, found before this shipped, not after**: the terminal renderer's Tools column
202
+ started annotating each tool name with its severity (`read_file [low]`), and a smoke test
203
+ against a real enumerated server showed the `[low]`/`[high]` suffix silently vanishing —
204
+ Rich's `Table` parses plain cell strings as its own markup language, and `[low]` isn't a
205
+ recognized style tag, so Rich absorbed it instead of erroring. Worse: a tool name or
206
+ enumeration-error string containing real markup syntax (`[/bold red]`) didn't just render
207
+ oddly, it **crashed** the renderer outright (`rich.errors.MarkupError`) — confirmed by
208
+ deliberately reverting the fix and re-running the test suite before shipping, not just
209
+ reasoned about. Tool names and error messages come verbatim from a live, untrusted third-party
210
+ server's own response (DEC-02's whole point), so every dynamic value reaching Rich now goes
211
+ through `rich.markup.escape` — server-config strings and, especially, anything sourced from a
212
+ live server's own output. Regression tests plant literal `[...]` sequences in a tool name, a
213
+ server name, and an enumeration error to prove none of them can vanish or crash rendering
214
+ again.
215
+
216
+ Verified end-to-end for real: `charter scan --enumerate` against the toy server, confirming
217
+ `read_file [low], write_file [high]` (or the equivalent ANSI-colored form) actually appears in
218
+ both the terminal table and matches `charter.lock`'s own `capabilities`/`severity` fields.
219
+
220
+ **Slice 3 (Session 16): markdown/SARIF renderers, the capability classifier's golden corpus,
221
+ and the CI accuracy gate.**
222
+
223
+ **Scope calls, same style as Session 15's:** no differ yet — `specs/charter.md`'s FR-006
224
+ "capability diff" PR comment needs one, and Session 15's own note already deferred that ("git's
225
+ own diff on the committed `charter.lock` is the review mechanism" until a real one exists);
226
+ building it belongs with the Action work in Session 17, where it has an actual consumer. So
227
+ `render_markdown`/`render_sarif` (`src/charter/render/markdown.py`,
228
+ `src/charter/render/sarif.py`) show the **current scan snapshot** — the same content
229
+ `render_terminal` already shows, reformatted — not a before/after. No JSON renderer either:
230
+ spec §6 lists `markdown | sarif | json`, but `charter.lock` (schema v3) already *is* the
231
+ canonical JSON artifact; a second one would just duplicate it. `charter scan --format
232
+ {table,markdown,sarif}` selects the display format (default `table`, unchanged); `charter.lock`
233
+ is still always written regardless.
234
+
235
+ **SARIF locations point at the server's config source**, not the tool — `server.source_file`/
236
+ `server.source_line`, since a live-enumerated tool has no line of its own to point at (DEC-02:
237
+ it comes from a `tools/list` call, not static text). Severity collapses charter's four levels
238
+ into SARIF's three real ones the same way ebb's own `render/sarif.py` does:
239
+ `LOW→note, MEDIUM→warning, HIGH→error, CRITICAL→error`. A tool matching more than one
240
+ capability produces one SARIF result per capability, same "don't under-report" stance
241
+ `capability.py`'s own docstring takes for `Classification.capabilities`. `--format sarif`
242
+ without `--enumerate` is not an error — it prints the full rule catalog with zero results,
243
+ since there's genuinely nothing to report yet.
244
+
245
+ **The golden corpus** (`tests/golden/<capability>/<NN_name>/tool.json`,
246
+ `tests/golden_manifest.yaml`, `tests/unit/test_golden_corpus.py`) scores the rules-only
247
+ classifier per capability group — never pooled into one aggregate, same "an average must never
248
+ hide one broken group" discipline `apps/ebb/tests/unit/test_golden_corpus.py` established —
249
+ against the same `PRECISION_FLOOR = 0.95` / `RECALL_FLOOR = 0.85` every other product in this
250
+ suite uses. 30 fixtures across six groups (the five `CapabilityClass` values plus `unknown`),
251
+ grounded in real tool definitions from the actual official filesystem/GitHub/Slack/Puppeteer/
252
+ Brave-Search MCP servers the taxonomy's own changelog already cites — not invented vocabulary.
253
+ Every group with an expected capability carries at least one pure-decoy fixture (a tool that
254
+ shares surface vocabulary with a real match without containing the actual matched pattern —
255
+ e.g. `readiness_probe`, which contains "read" but not the `read_`/`_read` substrings the rules
256
+ actually check for), enforced by its own test so a rule that over-matches can't hide behind
257
+ fixtures that only ever offer true positives. `golden/unknown/` is exempt from that
258
+ requirement: every one of its fixtures is already a negative case by construction.
259
+
260
+ **A real precision bug, found by the corpus before it shipped, not after**: `get_pull_request`
261
+ (a real GitHub MCP tool) was classified `network_egress` in addition to the correct `read` —
262
+ network_egress's bare `"request"` name pattern matched the substring "request" inside
263
+ "pull_**request**", nothing to do with an actual outbound call. The same bug would have hit
264
+ `create_merge_request`, `approve_service_request`, or any other tool whose name merely contains
265
+ the English word "request". Fixed in `capability_taxonomy.yaml` v2 (changelog entry in the file
266
+ itself): bare `"request"` replaced with the specific compound forms a real HTTP-client-shaped
267
+ tool actually uses — `http_request`, `api_request`, `make_request`, `send_request`,
268
+ `web_request`. Confirmed by re-running the full corpus before and after: 0.800 precision on
269
+ `read` before the fix, 1.000 after, every other group unaffected.
270
+
271
+ Verified end-to-end for real: `charter scan --enumerate --format markdown` and `--format sarif`
272
+ against the toy server, confirming real tool names/capabilities/severities appear in both, the
273
+ `--format sarif` SARIF log validates as well-formed JSON with the expected rule catalog, and
274
+ neither renderer leaks a planted fake secret value from a config fixture that has one (only the
275
+ existing `AIRTABLE_API_KEY` *name*, never its value — DEC-06, same check Slice 1 and 2 each
276
+ ran). `make accuracy-charter` reports 1.000 precision / 1.000 recall on every scored group.
277
+
278
+ **Slice 4 (Session 17): the capability-drift gate, the composite Action, PR comment, and
279
+ publish-prep.**
280
+
281
+ **The gap this session closes:** unlike ebb (`--fail-on`) and telltale (regression-vs-merge-
282
+ base), charter's `scan` had no failure condition at all through Session 16 — a naive Action
283
+ would only ever comment, never fail a PR, undercutting the product's own pitch ("fails CI when
284
+ an update quietly expands what the agent can reach"). `charter scan --base <ref>` now exits 1
285
+ when a new server appears, or an existing tool's capability severity increases, since the
286
+ merge base — never an absolute threshold, mirroring telltale's own regression-only philosophy
287
+ exactly (`src/charter/drift.py`).
288
+
289
+ **The baseline is the committed `charter.lock`, not a live re-scan** — DEC-04's own framing
290
+ ("`charter.lock` *is* the review mechanism") drives this directly: `charter/git_ref.py`'s
291
+ `show_file_at_ref` reads the lock file's content at the merge base via `git show`, never
292
+ launching whatever third-party MCP server versions existed at an old commit the way telltale's
293
+ worktree-based `--base` has to (it has no single lockfile-shaped artifact to diff against).
294
+ `toplevel`/`merge_base`/`GitError` moved to `packages/keel/src/keel/git_ref.py` this session —
295
+ pure git plumbing telltale's own `git_ref.py` already had, extracted because charter needed
296
+ the identical thing (same "genuinely identical, not merely similar" bar Session 15 used to
297
+ extract `keel.collect.line_tracking`); telltale now re-imports them, keeping only its
298
+ telltale-specific `checkout_ref`.
299
+
300
+ **A real, honest limitation, not a silent one**: tool-level drift (`NEW_TOOL`/
301
+ `SEVERITY_INCREASED`) is only computed when *both* the current scan and the baseline commit
302
+ were `--enumerate`'d — a server that wasn't enumerated on either side has nothing honest to
303
+ compare (not "no tools", not "no drift", just unknown), the same null/empty/populated
304
+ discipline `lockfile.py`'s own `tools: null` already encodes. `NEW_SERVER` drift has no such
305
+ limitation — a server's mere presence in the config is real evidence regardless of
306
+ enumeration.
307
+
308
+ **`apps/charter/action.yml`** (subpath, not the repo root — ebb already claimed that):
309
+ `enumerate` defaults to `"false"` (DEC-02 extended to the Action layer — launching third-party
310
+ servers in a shared CI runner is still a conscious opt-in, not a default; the documented
311
+ consequence is that only `NEW_SERVER` drift fires out of the box), `base` defaults to the PR's
312
+ base SHA (the gate itself *does* default on, unlike `enumerate` — matching telltale's own
313
+ `base` default). Steps mirror `apps/telltale/action.yml` exactly: scan with `--format
314
+ annotations` → render `--format markdown` for the sticky PR comment → upsert by marker
315
+ (`apps/charter/scripts/upsert_pr_comment.sh`, a byte-identical copy of the generic root-level
316
+ script) → fail on the captured exit code.
317
+
318
+ **Annotations, not a SARIF-upload step, for inline PR feedback** — verified live against
319
+ GitHub's own code-scanning docs before deciding: `github/codeql-action/upload-sarif` needs
320
+ `security-events: write` and, on private repos, GitHub Advanced Security/Code Security enabled.
321
+ Workflow-command annotations (`src/charter/render/annotations.py`, same
322
+ `::error file=...,line=...::` mechanism and escaping helpers ebb's and telltale's own
323
+ `render/annotations.py` use, duplicated on purpose) need neither — zero extra permissions, works
324
+ on every plan — so that's the composite Action's default. SARIF (`render/sarif.py`) stays
325
+ available for anyone who wants to wire their own upload step; it was not extended with a drift
326
+ section this session (still a snapshot renderer, Session 16's own call) since annotations
327
+ already cover the narrative-diff job.
328
+
329
+ **Publish-prep, matching `docs/adr/0002-ebb-public-mirror.md` and telltale's own Session 16
330
+ boundary exactly**: `.github/workflows/charter-release.yml` (tag prefix `charter-v*`),
331
+ `scripts/mirror-workflows/charter-{ci,release}.yml`, and `scripts/sync-charter-mirror.sh` were
332
+ added, but the script was **not run against a real repo**. `apps/charter/LICENSE` (Apache-2.0,
333
+ "the foretop authors", byte-identical to ebb's/telltale's own) is new; `license-files =
334
+ ["LICENSE"]` added to `pyproject.toml` and verified for real by building the wheel and
335
+ confirming `LICENSE` lands in `dist-info/licenses/` — telltale's own Session 16 found this
336
+ missing the hard way, so it was checked directly this time rather than assumed. The
337
+ `foretop-charter`/`charter` dual script entries telltale's Session 16 had to add were already
338
+ both present in charter's `pyproject.toml` from Slice 1 — nothing to fix there.
339
+
340
+ **Still the user's job, not this session's** (`NEXT_STEPS.md` Track A, same split every
341
+ predecessor used): creating the `foretop-dev/charter` repo, actually running
342
+ `sync-charter-mirror.sh`, creating the `foretop-charter` PyPI project, configuring its trusted
343
+ publisher (owner `foretop-dev`, repository `charter`, workflow `release.yml`, environment
344
+ `pypi`), cutting a `charter-v0.1.0` release, and Marketplace submission. The `foretop-keel`
345
+ workspace-only dependency problem `packages/keel/README.md` used to document as unresolved for
346
+ this mirror (and ebb's/telltale's) is fixed as of a later session — see that file's own
347
+ "Packaging note" — `scripts/sync-charter-mirror.sh` now vendors the exact `keel` submodules
348
+ charter imports (`git_ref.py`, `collect/line_tracking.py`) into the mirror at sync time.
349
+
350
+ Verified end-to-end for real: a real git repo fixture (commit a `charter.lock` with one
351
+ server, branch, add a second server) drives `charter scan --base <branch>` to exit 1 with a
352
+ `NEW_SERVER` finding visible in both `--format annotations` and `--format markdown`; an
353
+ unrelated change on the same branch exits 0; an unknown `--base` ref exits 2. `make check`
354
+ passes across `packages/keel`, `apps/telltale`, and `apps/charter` together, confirming the
355
+ `keel.git_ref` extraction didn't regress telltale.