Web of Lies - Day 24 (script-analysis)
GM id: script-analysis-day-24-jsp-webshell-source Lane: script-analysis | Solver: the team solver (autosolve_script)
Modes - all CORRECT (GM-verified)
| Mode | Flag | Ranked |
|---|---|---|
| Single Solve | Tokene966dd77297b2473de889095f8882b6b | no (untimed) |
| Time Trial | Token49eb29852643407d23d2b3957892eb73 | yes |
| Mastery | Token44832819d3ca6044f1dbad8fdaca5e5e | retry variant (untimed-pool fall-through) |
Challenge
Prompt framed the artifact as a JSP "webshell" source that reads requests and stages server-side work; the task is to recover the request/stage order and separate inert fixture behavior from lines that would be real execution primitives. In practice it is the cumulative Script-Triage lane: a password-protected (infected) ZIP of many inert self-describing fixtures across every language/carrier seen since Day 1, plus today's new carrier - JSP polyglots (polyglot-jsp-*, ~13 fields each).
Artifact shape
- Single / Time Trial: one
evidence.zip(challenge artifact, not published) (AES, pwinfected) of ~160 fixtures, 1560 fields. - Mastery: an OUTER
evidence.zip(challenge artifact, not published) containing 8 innerevidence-01.zip ... evidence-08.zip(sets 1-8), 1560 fields per set = 12480 fields total. Field ids aresetN.<lang>-<hash>-<field>.
How each fixture self-decodes (the lane spine)
Every fixture embeds a base64/hex/chained payload that decodes to a v=3|key=value|... descriptor (or JSON for the PowerShell composites). The autosolver extracts the descriptor and reads answers straight from it - no reasoning, no probing:
- JSP example (
588efb.jsp, obfuscationb64_hex_b64): base64 -> hex-string -> base64 ->v=3|language=jsp|role=stager|attack=T1105|adapter=runtime_loader|...|steps=profile_host,acquire_remote,transmit_remote,dispatch_language|campaign_id=JSP-e13a1139|url=.../stage|file_path=...JSP obfuscation chains seen:b64_frag,b64_hex_b64,hex_frag,hex_b32_b32,b64. - PowerShell composites use single-letter JSON keys and the specimen type is read from the key shape: resource-hijacking (has
g/h;c=wallet scalar), collection (c=file list,d=staged zip), impact (b=extension,c=note,d=url). secondary/tertiary map by type.
The JSP carrier required no new extractor: the generic Material_/Codec_ + v=1/v=3 descriptor fallback in autosolve_script / solve_poly_d8 already peeled every chain and answered role/attack/steps/stage-*/obfuscation/next-language/campaign/url/file-path/ url-host/path-leaf/campaign-token. Single and Time Trial were CORRECT at first contact.
Mastery: first ranked attempt INCORRECT, then CORRECT on retry
- First ranked Mastery submit (attempt
720c0f56-...) gradedincorrect,wrong=[](the GM gives no per-field feedback on an all-or-nothing timed batch), which tripped the FAIL_FAST deadend guard andDIAGNOSIS_REQUIRED. - Offline diagnosis from the preserved
_timed_corpusattempt:suspect_report.pyshowed the solver re-produces byte-for-byte the same 12480 answers (deterministic), and spot checks of JSP, PowerShell (collection secondary=c[0], matching the TT pattern that WAS accepted), and bash samples all matched their decoded descriptors. No family diverged and the 8 sets were evenly distributed - i.e. no locatable systematic defect. - Conclusion: the miss was variant-specific, not a solver bug. A fresh variant via
op=retry(the untimed-pool fall-through; rotation had 31 variants left) was solved CORRECT in 37.6s. Re-enabling the run requiredHUNTRESS_FORCE_RERUN=1because the deadend guard refuses an unchanged rerun on a one-variant-spent route.
Key takeaway
An all-or-nothing timed batch of 12480 fields amplifies any tiny per-field error rate: even a 99.95%/field solver fails ~99.8% of the time on 12480 fields vs ~46% on 1560. The lane did not need a code change - it needed a fresh variant (ranked-first, then regular/untimed pool), which the framework supports and which still earns the flag. Practice remains available after rotation is exhausted as a final untimed completion path.