Point the Updates links at updates.php
The wiki still linked to the retired updates.html on every page via the
sidebar, plus once on Home. That stale page is what search engines and
LLM crawlers were finding and quoting, so the project looked like it had
not shipped anything since 30 May 2026.
d98f639
Issue trackers: workflow rules work the same for both (#967)
Ed asked whether the workflow module can drive Azure DevOps like it does
Jira. It can — you pick a CONNECTION, so the tracker follows from that
and nothing about writing the rule differs. The page now says so, since
"raise it in Jira" implied otherwise.
Added, because they are the two ways a rule can be written and still be
wrong:
- leave Project and Issue type BLANK and the connection's mapping
decides, which sidesteps the next point entirely;
- ⚠️ if you fill them in they mean different things per tracker — Jira
wants the key, Azure DevOps the name — and Azure DevOps issue types
come from the project's template, so a rule hard-coded to Bug works
against an Agile project and fails against a Basic one in the SAME
organisation.
Also noted that the tracker triggers can branch on which tracker fired
them, so per-tracker rules are possible rather than one rule having to
serve both.
Neutralised the last two Jira-shaped paths in shared sections (the
mapping screen's location, the dry-run note).
Verified no dead in-page anchors.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
c400994
Issue trackers: make the analyst page genuinely cover both (#964)
Ed asked "what should the project key be if I send it to devops" — and
that question was the review. The page defined project key as "the KAN
in KAN-6", which is not merely Jira-flavoured wording but ACTIVELY WRONG
for Azure DevOps, where there is no key at all and the project NAME is
the identifier.
The connector page shipped with the connector; this is the analyst half
catching up properly rather than having a DevOps preamble bolted onto a
Jira walkthrough.
- "Which project?" is now a two-row table, because the answer genuinely
differs: Jira wants the KEY, Azure DevOps wants the NAME, and getting
it wrong produces a permission-shaped error that reads like bad
credentials.
- "Which issue type?" gains the per-PROCESS table. ⚠️ Two projects in
ONE Azure DevOps organisation can disagree about whether Bug exists —
Basic has no Bug at all — so "raise it as a Bug" is not a safe default
the way it reads on Jira. Also documents that a blank type becomes a
Task, because Task is the one type in every process.
- The connection table is now Jira | Azure DevOps side by side, which is
where the missing email field and the extra Resolved setting become
obvious rather than needing prose.
- Neutralised the shared prose that quietly assumed Jira: the scheduled
check, the multi-company rule, the never-blocks-the-ticket note and
the comments section all describe behaviour that is not Jira's.
Kept Jira-specific what genuinely is — the Atlassian token steps are
scoped with a marker rather than generalised into uselessness.
Verified: no dead in-page anchors after renaming "Comments from Jira",
no dead page links.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
749d0be
Issue trackers: Azure DevOps connector, and the contract test it settled (#964-966)
New page: Azure-DevOps-Connector-Developer-Guide, the sibling of the Jira
one. Issue-Trackers gains Azure DevOps setup and the Resolved setting;
the two guides that PREDICTED this work now record how it turned out.
The engine guide committed to a falsifiable claim before Jira shipped:
adding connector #2 must touch nothing but the provider, one registry
row and its settings partial. It held, and the page now says so with the
three things that did change for everyone:
- settings_fields, and why it is separate from credential_fields — a
dropdown among the credential boxes inherits their blanking rule and
silently resets on every save;
- integrationsAbsoluteUrl(), because the ticket link every connector
puts in an issue was built from BASE_URL, which is a path, so it 404'd
from inside any tracker and had never worked;
- $target may be keyed either way — core's neutral `issue_type` or the
provider's own word — because reading only one silently creates the
wrong kind of work item.
The Jira guide's "if you are writing the next connector" list is now
marked as validated, with two of its decisions reclassified as Jira-ONLY
now there is something to compare against:
- relative JQL dates exist because Jira reads an absolute date in the
USER's timezone. Azure DevOps honours an explicit UTC Z. Copying
either across is a bug, and both are right for their own tracker;
- flavour-dependent author_identity is an Atlassian quirk, not a general
requirement. What IS general is that it match testConnection().
What the new page records that the code cannot say for itself:
- ⚠️ which field holds the body depends on the work item TYPE. A Bug's
form shows Repro Steps and not System.Description, so writing the
description to the obvious field returns 200 and produces a work item
that looks empty to the developer who opens it;
- 🔑 five state categories, not four, and why the fifth is a per-connection
SETTING rather than a mapping — plus the live vindication of "never
branch on a status name": "Resolved" is category Resolved on a Bug and
InProgress on a User Story, same project;
- ⚠️ timePrecision=true is a query parameter, not a body field, and the
tempting fix for the error it throws silently re-reads a day of comments;
- ⚠️ echo suppression must use createdBy.id — the two endpoints report
DIFFERENT descriptor formats for the same person, so comparing those
never matches and every comment we write comes back as a developer's;
- ⚠️ attachments are two steps and step one attaches the file to nothing;
- ⚠️ a rejected token answers 203 with a sign-in page, not 401 with JSON.
Also corrected two stale rows in the analyst page's "not there yet"
table: attachments and "tell the requester when it is done" both shipped
weeks ago and were still listed as missing.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
a43d94e
Forms: lookup fields — analyst page + developer guide (#962/#963)
New page: Forms-Lookup-Fields-Developer-Guide. The security surface
earned it its own guide rather than another section in the sections &
conditional-logic one — three gates, a scoping rule whose NULL means the
opposite of Knowledge's, and a bug that only showed up under a live run.
Forms.md gains a Lookup fields section (and the type count moves from
nine to ten, which the page had been getting right by hand).
What the dev guide records that the code cannot say for itself:
- Why one type with a source, not one per source. Same argument as the
date field's mode, and it will come up again for the next type:
field_type cannot change once a field exists.
- The registry IS the SQL whitelist — those column names are
interpolated, so they may only ever come from the constant. Four of
five column sets assumed while building were wrong, and a wrong column
name renders as "no matches" rather than an error, so a source can be
broken and look merely empty.
- Why contracts and software are absent, with the specific schema gap
each is waiting on.
- The three meanings of $tenantIds, since null / [] / [ids] are all
legitimate and confusing two of them is silent.
- ⚠️ NULL tenant means "the Default company's" here, the opposite of
Knowledge — and on a single-company install every row is NULL, so the
clause handling it is not an edge case, it is the feature working.
- The #963 post-mortem, because the generalisable part is not the fix:
a function_exists() guard around a security rule degrades silently,
and not necessarily toward the safer behaviour. Here it degraded
toward refusing legitimate answers.
- That adding a field type means five places, one of which is PROSE —
and that the prompt had already drifted, omitting datetime and section
for months without anything failing.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
4799f0b
Issue trackers: sending an attachment after escalation (#961)
Records why it is a button rather than automatic, and the guards on an
endpoint that takes two ids from the browser — chiefly that the
attachment must belong to that ticket, or any file in the system could
be posted to any issue.
546ac3c
Issue trackers: attachments proven live, and refresh the stale header
A real 83 KB PDF went from a ticket to KAN-6 and Jira reported it back
byte-identical, so 7g is no longer 'unproven live'.
The page header had drifted two releases behind: it still said tracker.*
triggers did not exist and quoted 252 assertions. Now #945-960, 288
assertions, 12/17 parity, with what is genuinely still missing listed.
3d8805b
Issue trackers: attachments (#960)
Gains 7g: why inline images are excluded (and what that costs), the
size/count rules, why a failed upload must never fail the escalation, and
the three Jira specifics that make an upload work — the no-check token, the
field named `file`, and a random boundary.
Also records a test lesson: asserting the bytes AROUND a sanitised
filename fails on correct output. Assert the value.
3a44bad
Time-Based Triggers: these had never actually fired (2026-08-02)
The page described the design correctly; the code did not match it. A
duplicate sla_format_minutes() fataled the SLA cron on invocation, so
sla.warning and sla.breached had never fired since the feature shipped
on 17 May — with two further wrong-column bugs hidden behind it.
Recorded here because the page is what someone will trust when adding
the next one: a trigger being registered, documented and visible in the
editor proves nothing about whether it fires. Make it fire.
8a2d9f8
Issue trackers: tracker.* triggers (#958), and why they are not time-based
Engine guide gains 7f: the three triggers, that each carries the ticket,
the aliased columns in the payload query that are silent when wrong, and
the comment loop that terminates only because send_note_to_tracker
records what it pushes.
Time-Based Triggers gains a "a cron is not the same as time-based"
section, because reusing the emissions ledger for these is the obvious
instinct and it is wrong — a fingerprint on current state would swallow
a genuine third transition.
8445e85
Issue trackers: separate the two "no change to the tickets module" claims
Escalation genuinely added nothing to the tickets module and that still
holds. The module as a whole stopped being untouched at #954, because
reading comments back means rendering them — get_notes.php, inbox.js,
inbox.css and tickets/index.php are all in it now.
Worth pinning because the stale version is exactly the sort of line that
ends up in published copy.
14378a5
Issue trackers: the pretty-URL relative-path trap on the help page
Three traps worth naming, all hit within minutes of adding the page: a
pretty URL deeper than its file breaks relative asset paths (use
BASE_URL); $path_prefix must stay filesystem-relative because the system
header requires through it; and a System page must use the system
header, not the root one, or it renders the ticketing nav.
71222c0
Issue trackers: the grouped routing rules and the in-app setup guide
Records why departments and companies are never one flat list (the
grouping is what makes precedence visible), and documents help.php —
including that its rewrite must precede the provider rule.
29b767e
Issue trackers: document mapping (#957)
Engine guide gains 7e: the precedence rule and why it is that way round,
why priority has no wildcard, the narrow rejected-priority retry, the
namespaced local_key, and two traps — the column is ticket_type_id not
type_id, and a shared service must not assume it owns the transaction.
Analyst page gains a Mapping section in plain terms, and the Jira page
records that mapping needed no connector change beyond listPriorities(),
which is the abstraction doing its job.
c5aee9b
Add a Scheduled tasks page; bring the tracker dev guides current
New page: Scheduled-Tasks. Ed asked for one place covering every cron
job / Windows task FreeITSM needs, because "you must schedule something
or comments never arrive" was buried in the Jira page. Covers all four
jobs, what silently breaks without each, Windows and Linux commands, how
to run one by hand, and how to tell whether they are working.
Also documents what is NOT a scheduled task, so nobody goes looking for
a job that does not exist or schedules the InTune workers, which the app
starts itself. Notes that mail is fetched by the browser, so nothing is
collected overnight with every browser closed.
Windows gotchas that cost time today: /ru SYSTEM tasks are invisible to
a non-elevated session, 0x41303 means "not yet run", and the escaped
quoting fails if a path has spaces.
Linked from Home, the sidebar, and five places in Issue-Trackers.
Dev guides brought current for #954-956: assertion count 227, range
#945-956, the pipeline diagram no longer shows the removed author
guard, and a new note on inbound_enabled — a column that shipped in V1
with no writer, so the feature it gated could never be turned on. That
is a trap worth naming for the other columns V1 shipped early.
06e4b2b
Issue trackers: why author-based echo suppression was removed (#956)
Records the failure and the reasoning, at length, because the idea is
tempting enough that someone will reach for it again: "authored by our
account, therefore our own echo" assumes a service account that never
types, and the token owner is usually a human who comments.
Also notes the uncomfortable part — the suite was green and contained
assertions pinning the wrong behaviour, because they shared the design's
assumption. Only a live account found it.
fb7f808
Issue trackers: say what "the check" is, and where the tickbox lives
Both raised by Ed while setting this up for real, and both were fair.
"Run the poll" was jargon that meant nothing to a user. The page now
explains the check in plain terms — FreeITSM asks Jira rather than Jira
calling us — says that nothing on the page works until it runs, and
shows how to run it once by hand for testing.
The Accept updates tickbox is inside the connection, not on the list
page, so the page now spells out the pencil-then-scroll route, notes the
token box can stay blank, and adds both to troubleshooting.
dbece7d
Issue trackers: author_kind, and why "Unknown" was the wrong answer (#955)
Records the four cases a note's author can be in, and that collapsing
them into one label discards information the row holds. Also notes that
delete_analyst.php reassigns nothing, so a departed analyst's notes are
the normal case rather than an edge case.
c027ddb
Issue trackers: comments coming back, and the join that hid notes (#954)
The engine guide gains 7d — the inbound pipeline, both echo guards, the
"first poll imports nothing" rule and why the comment map row is written
before the note (the unique key is what wins a race between two cron
runs, not the check before it).
The Jira guide gains 7: two calls rather than one per issue, why the JQL
date must be relative (an absolute one is read in the Jira user's
timezone, not UTC), why the ADF reader lives in the connector rather
than IssueDoc, and that author_identity must match testConnection() or
echo suppression silently stops working.
The analyst page documents the toggle, that imported comments are always
internal, and that enabling it does not bring back a backlog.
Also recorded: get_notes.php inner-joined notes to their author, so 10
of 17 notes on the dev install were invisible. Both dev pages now list
what forces an update, including the rules above.
de457a2
Tracker guide: correct the shipped range, point at the parity scoreboard
4603699
Issue trackers: the analyst page, and cross-links (#945-953)
The dev pair existed but the analyst page did not, so a user had no
page at all. Issue-Trackers covers setup (including the API token
living in the Atlassian account, not Jira), raising an issue, reading
the pill, automating it with a workflow rule, and an honest 'what is
not there yet' — comments coming back is named as the biggest gap
rather than left to be discovered.
Also: sidebar entry, and both dev pages now point at it.
ebe7f6a
Jira guide: the removed search endpoint, and real-Jira proof (#953)
Cloud retired /rest/api/3/search for /search/jql; DC keeps the old one.
Recorded with the lesson that generalises — a connector's correctness
depends on someone else's API, so when a call starts failing, suspect
the endpoint before the logic.
Status banner now records that the whole journey has run against a live
Jira Cloud site.
6aa2531
Tracker guide: the poll cron (#951)
Records that the poll is currently the ONLY way status changes, the
per-connection interval, stamping on failure, and the fetchIssues()
distinction — swallow one bad chunk, but rethrow when every chunk
failed, because [] is indistinguishable from 'those issues do not
exist' and the poll would look healthy while the tracker was down.
aecdb3f
Tracker guides: the ticket side (#950)
§7c covers the pill in the Links strip, the manual escalate and its
preview, integrationsBodyToText (and why strip_tags alone leaves a
stylesheet in the description), and the two traps — modals open with
classList.add('active') not style.display, and inbox.js?v= must be
bumped. Both present as 'the button does nothing'.
Header no longer claims 'engine only'; status now says raising works
but nothing comes back yet and the poll cron is missing.
943ec98
Bring the tracker guides up to #948 and #949
The pages documented up to #947 and then two commits shipped without
them, so §9 confidently said "not built" about the settings screen and
the workflow actions — both live. A wiki that is wrong is worse than
one that is missing, because someone trusts it.
Added: §7a the workflow surface (the two actions, the connection
lookup, argBool, and why dry-run safety lives in runInner rather than
the handler), §7b the settings screen (the folder-level .htaccess
pattern, one shared provider.php, and the four traps — .form-row being
display:flex in inbox.css, full-width needing BOTH no max-width and no
auto margin, showConfirm/showToast over browser dialogs, and closest()
once a button contains an SVG).
Updated: the file table, the §9 status table, §10 to cover the provider
registry that keeps a new tracker cheap, the contract for
pollChanges()'s watch list, and a note that the service is
self-contained because it once was not and fataled on decryptValue().
Recorded the miss itself in "keeping this page honest" — same commit,
not the next one.
1b36d0d
External issue trackers + Jira connector developer guides (#945-947)
Two pages, split on the line that matters: the engine page is the
contract every tracker must satisfy, the Jira page is the worked
example. That split is the point — whoever writes the GitHub or Azure
DevOps connector needs to know which decisions were forced by the
contract and which were just how Jira happens to work, and both pages
say so explicitly.
Also cross-referenced from Multi-Tenancy-Developer-Guide: tracker
connections added to the connection column of the three-meanings-of-NULL
table, and the outbound twin of the membrane rule recorded next to the
inbound one.
Both pages carry a "keeping this page honest" section listing what must
be updated in the same commit as a code change.
d00486b
Forms: document date / time fields (#944)
Forms.md — nine question types, a "Date and time fields" section covering the
three modes and why it is one type with a setting rather than three types, the
is_after / is_before row on the operator table plus a note that the menu narrows
per trigger type, and the timezone warning.
Dev guide — new §3b on date fields: the mode, the stored formats, why these are
NAIVE values and must not be converted, why the operators compare as strings
(ISO-8601 lexical order is chronological order), and opsFor() narrowing the menu
without narrowing the API. Also: config must be MERGED not replaced (an earlier
buildRulesForSave dropped date_mode), the system prompt is part of the feature
and drifts silently (section was whitelisted in #940 but never described to the
model, so it was never emitted), and adding datetime was the first real test of
the deliberate three-renderers decision.
Every app file named verified to exist; every wiki link resolves.
87a26ef
Forms: document sections, conditional logic and field identity (#940-943)
Forms.md — nine types (section added), a Sections section, a Conditional
questions section with the operator table and the three deliberate behaviours
(hidden is not required; hidden answers are not stored; conditions look
backwards only), and what "removed" means on a submissions column.
New Form-Sections-and-Conditional-Logic-Developer-Guide.md — the colour-keyed
file table, why field identity had to be fixed first (with the old positional
sync quoted), the three identity spaces (id / "idx:N" / _key), why a section is
a row rather than a table, the backwards-only rule as the thing that makes
cycles impossible by construction, the two evaluators and the one case table
that keeps them honest, soft delete, version-fork remapping, why the three
renderers were deliberately NOT merged, the AI-apply trap, and the upgrade note
for REST clients.
Every app file named verified to exist; every wiki link resolves.
9801e78
Knowledge Assistant: user page + developer guide
New pair for #939. The user page leads with why the assistant is allowed to
refuse, since that is the design decision the whole feature rests on. The
developer guide carries the colour-keyed file table and documents the three
things nobody would guess: the buffered verdict line, overlap-coefficient
similarity with a two-token floor (not Jaccard), and single-linkage clustering
- the last two both forced by tests/knowledge-gaps failing.
Also a section on Knowledge.md and both pages in _Sidebar.md.
35385e5
New: Asset scanning — Developer Guide (the deep dive)
The scanning journey was spread across three pages and mostly implied.
This traces it end to end: how a printed square of ink becomes an
updated database row.
Twelve sections, the substantive ones being:
- Three scanning surfaces (native camera / in-app scanner / USB gun)
and why none of them is redundant.
- The public base URL — the single most consequential line in the
feature. Why it must not come from the request, the localhost
banner, the sub-folder double-append that would otherwise be
printed onto physical labels, and the two tunnel traps (a rotating
free ngrok URL, and its browser-only interstitial that curl won't
show you).
- The resolve flow, including why three regexes deliberately disagree
(routing is permissive, validation is strict, dispatch is neither).
- The write path: one door, shared with the desktop editor, so a
phone edit gets the same audit trail and warranty sync — and why
the no-op skip is load-bearing both in production and in tests.
- The security model: the token is a name, not a password, and the
unknown-vs-another-company answers are identical on purpose.
- Two mobile strategies in one module — mobile-FIRST scan pages
versus the mobile-ADAPTED module — and the five-step hinge where
they meet, which is what makes a scan land on the detail pane with
the device Back button working.
- How to test a camera headlessly (Y4M fake camera) plus the four
traps hit doing it.
- Failure modes, each mapped to its actual cause.
Section 1b of the labels guide now points here instead of duplicating
it; cross-linked from both QR pages, Mobile: Assets, and the sidebar.
18cb14b