Repository navigation
Glossary
The canonical vocabulary of this wiki and of the kit, in the six groups the documentation uses. Terms in bold are the canonical forms enforced by the run's terms.json; the right column names the rejected near-synonyms where they exist — one concept, one term, across every page. Definitions are traced to the source at 7e17a11 or to this run's tests.
- Doctrine
- Phases and process
- Tools and mechanisms
- Artefacts
- GitHub wiki mechanics
- Sandbox and operations
- See also
| Term | Definition | Not this |
|---|---|---|
| the kit | this repository/toolchain as a whole: doctrine docs + 11 Python tools + 3 shell scripts + templates + examples |
the tool, the framework, the product (ambiguous with individual tools) |
| the one rule | a claim enters the wiki only if traced to a source line, or reproduced by executing the software |
golden rule, core principle
|
| subject | the repository being documented |
target repo; project (ambiguous — could mean the kit) |
| run | one end-to-end execution of the four phases |
session, pass (for the phases) |
| reference run | the completed MD.Viewer run the kit was extracted from: 22 files, 73,013 words, 154 alerts, 27 mermaid diagrams, 1,687 table rows, 341 wiki-links, 0 QA errors | "original run" |
| field run (run 2) | the MD.Scribe 3.1.0 run behind the 1.1.0/1.2.0 hardening: 23 pages, 41,623 words, 249 empirical checks, 14 verified doc-vs-code discrepancies | — |
| verification channel | either half of the one rule: (a) source tracing or (b) execution | — |
| provenance | the source discipline on every fact: code:file:line | readme | git | test:name | web:url
|
— |
| doc-vs-code discrepancy | a documented claim the executed software contradicts; recorded, corrected in the wiki, and banned from returning | — |
| drift finding | a measured behaviour that no longer matches an expectation recorded from an earlier build (this run found two in the current MD.Viewer) | — |
| Term | Definition |
|---|---|
| Phase 1 — research (~40% of effort) | bootstrap → clone → repo_scan.py → run and probe the subject → the five research files |
| Phase 2 — wiki architecture (~3%) | page tree in PROGRESS.json; ban.json from sources §D; terms.json from the glossary |
| Phase 3 — writing (~45%) |
new_page.py --init scaffold → fill pages (hardest first) → fix_toc.py per rewrite |
| Phase 4 — QA and delivery (~12%) |
--self-test → --strict to zero errors → re-run documented commands → publish.sh → REPORT.md
|
| phase boundary | the moment state files are updated and a snapshot taken — not "at the end" |
| recovery point |
STATE.md / PROGRESS.json / REWORK-STATE.md / research/ — enough for a fresh agent to resume without re-deriving anything ("checkpoint" is acceptable in prose) |
| effort budget | the measured 40/3/45/12 split from the reference run |
| hand-over | the delivery package when pushing is impossible: prepared clone + verified bundle + PUBLISH.md
|
| Term | Definition | Not this |
|---|---|---|
| battery | a declarative set of probe requests: traversal (20 payloads), injection (26), headers (10 security + 4 leak), flags, disk |
test suite (reserve suite for the registered functions) |
| suite | a registered probe function (Probe, entry) -> dict: generic, php-mdviewer, or yours |
— |
| staged instance |
ServedApp's clean throwaway copy of the subject on a verified-free port, killed on exit |
test server, sandbox app
|
| verified-free port | a port bound by a probe socket before the server starts; a busy preferred port raises instead of drifting |
free port (weaker) |
| body marker | a substring asserted in a response body, because a status code alone can be meaningless (front controllers answer 200 for everything) | content check |
| evidence | JSON/Markdown artefacts recording requests, statuses, headers, excerpts — the reproducible backing for [tested] claims |
"proof", "output" |
| runtime floor | the subject's declared minimum runtime (composer.json → require.php, engines.node, go.mod); mismatch = exit 2, because probing an unbootable subject measures a corpse |
— |
| capability report | bootstrap's inventory of installed tools, stdlib modules and network egress | env report |
| scrubbed view | page text with fenced/inline code, [[wiki-links]], anchor targets and filenames blanked (offsets preserved) before ban/avoid/filler matching |
cleaned text |
| negation downgrade | a banned-phrase hit preceded in-sentence (~90 chars) by a negation/hedge becomes a WARNING — never a suppression |
suppression (it is the opposite) |
| exemption |
{phrase, file?, quote?, reason} entry making a legitimate hit report as INFO, visible forever, never silently dropped |
whitelist |
ban list (ban.json) |
phrases that must never appear, built from sources.md §D; affirmative forms only |
blocklist |
terms file (terms.json) |
canonical terms (counted, never failing) + avoided synonyms (errors) + _tuning_log
|
glossary file (the glossary is this page / the domain-notes section) |
| self-test | a tool's built-in proof that it can fail: qa 22 detections, fix_toc 8, evidence 3, session 8 (live PHP) | — |
| QA-clean by construction | scaffold output passes qa_wiki.py --strict unfilled (verified across three page lists, T8) |
— |
| negative control | a deliberately broken input proving a CI step fails when the defect returns (five named in the 1.2.1 verification) | — |
| Term | Definition |
|---|---|
| the five research files |
summary.md, architecture.md, domain-notes.md, facts.json, sources.md — Phase 1's output and Phase 3's input |
| sources.md §D | claims deliberately NOT made, with reasons; the most valuable section, because it becomes ban.json
|
| facts.json | machine-readable fact base; every value carries its source
|
| PROGRESS.json | page tree + groups + style contract + do_not_claim + QA numbers + delivery state |
| STATE.md / REWORK-STATE.md | orientation file / recovery artefact (resume recipe, environment rebuild, quirks, append-only log) |
| REPORT.md | the closing report: inventory, verification methods, doc-vs-code findings, sources, and the honest limits of the analysis |
| PUBLISH.md | the written push instructions: clone path, branch, SHA, remote, bundle, three options, never-force-push rule, post-publish checklist |
| bundle |
git bundle of the wiki commit — verified with git bundle verify; cloning one directly needs -b master (no HEAD symref, else an empty tree and vacuous checks) |
| Term | Definition |
|---|---|
| the filename rule | files are named with spaces (Security Hardening Guide.md); GitHub maps them to hyphenated URLs and resolves [[Page Name]] against that; hyphenated filenames kill every such link |
| GitHub slug | heading → anchor: lowercase; strip everything except alphanumerics/spaces/hyphens/underscores; spaces→hyphens; do not collapse repeated hyphens; dedupe -1, -2. Computed by tooling, never hand-written |
| chrome files |
_Sidebar.md and _Footer.md — underscore-prefixed, rendered on every page, not listed; the sidebar must link every page |
| alert | `> [!NOTE |
| fence nesting | an example containing a fence needs a longer outer fence (four backticks around three); the early-close bug renders the rest of the page as garbage |
| wiki repo |
OWNER/REPO.wiki.git, default branch master, created by GitHub only after the first UI page save |
| Term | Definition |
|---|---|
| PITFALLS |
setup/PITFALLS.md — 31 traps in four classes (A wrong-answers, B sandbox, C wiki, D process), ordered by time wasted; A1–A3 produce silently wrong results |
| A1 / stale server | a port collision makes the new server die while tests hit the old one — the worst offender, and the reason serve.py exists |
| B4 / path mangling | absolute workspace paths inside heredocs and script files can be rewritten or left literal by the harness; resolve paths at runtime inside scripts (this run's field discovery) |
| B11 / visibility loss | a long probe losing sight of parts of the workspace filesystem mid-run; the reason evidence.py writes with fallbacks |
| snapshot |
snapshot.sh's git commit of the whole workspace at a phase boundary, so a bad bulk edit costs minutes |
| D2 discipline | distrust your own checker once: verify one case by hand on mass failure; verify it can fail on mass success |
| terminal-less sandbox | no controlling terminal, closed stdin: no prompts anywhere, git push fails → bundles and hand-overs are the delivery model |
-
Configuration Reference — the schemas behind
ban.json,terms.json,PROGRESS.json - Tool Reference — every term's owning tool
- Writing Pages Guide — the style contract the glossary enforces
- Sandbox Survival Guide — the PITFALLS classes in operational context
- Home — wiki entry point
github-wiki-llm-writer 1.2.1 · MIT · Repository · Report an issue · Home · Glossary · FAQ · Troubleshooting
Facts in this wiki are verified against the source at commit 7e17a11; see Changelog and Versioning.
Home Getting Started Features and Verification Model Comparison with Alternatives
Installation and Setup Configuration Reference
The Four Phase Workflow Research and Probing Guide Writing Pages Guide QA and Audit Guide
Publishing and Delivery Security Considerations Sandbox Survival Guide Extending the Kit
Tool Reference Architecture Troubleshooting FAQ Changelog and Versioning Glossary
1.2.1 · MIT